Geotech Geol Eng

DOI 10.1007/s10706-015-9843-2

ORIGINAL PAPER

Critical Review of Thermal Conductivity Models

for Unsaturated Soils
Yi Dong • John S. McCartney • Ning Lu

Received: 11 January 2014 / Accepted: 6 January 2015

Ó Springer International Publishing Switzerland 2015

Abstract Although it is well established that heat saturation. A conceptual model based on soil–water
conduction in unsaturated soil depends on liquid retention mechanisms, is proposed to overcome the
saturation, there are several models available to pitfalls of the existing models and can be used to
consider the changes in thermal conductivity during establish quantitative thermal conductivity models for
drying and wetting. The key factors affecting thermal variably saturated soils in the future.
conductivity of unsaturated soil are evaluated through
a critical examination of these different models and Keywords Thermal conductivity  Heat transfer 
their development. Depending on the principles and Unsaturated soil  Soil–water retention
assumptions employed, these models are categorized
into three groups: mixing models involving series/
parallel elements; empirical models where thermal 1 Introduction
conductivity values at dry and saturated states are
used; and mathematical models based on phase Heat transfer in soil and rock presents a field of fertile
volume fractions. Experimental data for different soils research in which the thermal properties of soils are
are used to assess the quality of prediction for these used in a large number of geotechnical, geophysical
models. It is found that all the existing models do not and geo-environmental applications, including geo-
realistically account for pore structure or interface thermal energy resources (White 1973), radioactive
properties, and thus are not capable of properly waste disposal (Li et al. 2012), ground-source heat
predicting thermal conductivity as a function of liquid pump systems (Preene and Powrie 2009), energy piles
(Brandl 2006), geological CO2 sequestration (Ebigbo
2005) and recovery of natural methane gas hydrates
Y. Dong  N. Lu (&) (Cortes et al. 2009).
Department of Civil and Environmental Engineering, Soil is a multi-phase material consisting of solid
Colorado School of Mines, Golden, CO 80401, USA
e-mail: ninglu@mines.edu particles, gas and/or liquid. Thermal properties of soils
are not only determined by the intrinsic physical
Y. Dong
e-mail: ydong@mines.edu properties of each phase, but also affected by variation
of each phase. The thermal conductivity k of different
J. S. McCartney soil phases varies over two orders of magnitude
Department of Structural Engineering, University of (e.g., thermal conductivity of mineral particles
California, San Diego, 9500 Gilman Dr., La Jolla,
CA 92093-0085, USA kmineral [ 3 W/m K, thermal conductivity of water
e-mail: john.mccartney@colorado.edu kwater = 0.56 W/m K (at 0 °C), and thermal

123
 Geotech Geol Eng

conductivity of air kair = 0.026 W/m K) (Mitchell and ignored. Under these assumptions, the apparent or
Soga 2005; Yun and Santamarina 2008). However, the effective thermal conductivity of soil is often used to
thermal conductivity of dry soil kdry_soil is almost one describe the heat transfer capability of multi-phase
order of magnitude lower than that of the pure mineral materials.
solids, in most cases kdry_soil \ 0.5 W/m K, depending The heat flux q (W/m2) at steady state is proportional
on mineral composition and packing density (Farouki to the thermal gradient by the coefficient of thermal
1981). This implies that the air obstructs heat conduction conductivity k (W/mK), according to the constitutive
and heat conduction primarily occurs through the particle equation of Fourier’s law in one dimension:
contacts in dry soil. At the other end, the thermal dT
conductivity of water-saturated soil is between that of the q ¼ k  ð1Þ
dx
pure mineral and that of water, implying that replacement
of air with water provides a significant improvement in The rate of heat transfer in transient conditions is
the heat conduction through the soil mixture. The ordered equal to the heat stored within the medium and the rate
sequence of typical thermal conductivity values are of internal heat generation, given by the continuity
kair \ kdry_soil \ kwater \ ksaturated_soil \ kmineral. These equation:
observations, and many experimental studies (Brandon dq
and Mitchell 1989; Farouki 1981; Smits et al. 2010), ¼ qstored  qgenerated ð2Þ
dx
confirm that the varying water saturation in the soil
mixture system plays an important role in determining where the heat stored in the material is: qstored ¼
the bulk thermal properties. q  c  ðoT=otÞ; q (kg/m3) is the material mass density,
The purposes of this paper are: (1) to identify the and c (J/kgK) is the heat capacity. If there is no heat
controlling physical mechanisms for the thermal generation within the material, the qgenerated term
behavior of the soil at various unsaturated conditions, vanishes. Combining the above two equations leads to
and (2) to assess the existing models for their Fick’s second law:
predictability of thermal conductivity under varying o2 T oT
saturation conditions. ¼ D ð3Þ
ox2 ot
where D ¼ k=ðq  cÞ (m2/s) is the thermal diffusivity
2 Heat Transfer Mechanisms and Governing which reflects how fast heat will be transferred
Factors through a material. The negative sign indicates that
heat transfer occurs in the direction opposite to the
There are three heat transfer mechanisms in a material temperature gradient. Common values of thermal
medium: conduction through solids and liquids, con- conductivity, diffusivity, and specific heat of different
vection in fluids, and radiation (which does not require soils and their phase components are summarized in
a material medium). The most effective means of Table 1.
transferring heat in dry particulate materials is through The thermal properties of soils under isothermal
the solid contacts, while conduction through the gas conditions are governed by several particle-level and
phase and radiation have less relevant effects (Carslaw macro-scale factors: mineralogy, particle size, particle
and Jaeger 1959; Murashov and White 2000). Heat shape, packing geometry, stress level, water content,
transfer by convection plays an important role if the porosity, gradation, and cementation.
particle size D50 is larger than *6 mm permitting Mineralogy Solid minerals are the most condu-
fluid flow through the porous network (Yun 2005). In cive constituents in the air–water–solid soil system;
this paper, the scope of our assessment of thermal thus they define the upper limit of the thermal
conductivity models considers only heat transfer by conductivity. Soil comprised of different mineral
conduction. It is assumed that soils are under room substances has different thermal conductivity (e.g.,
temperature and under a small temperature gradient. quartz [ mica) (Gangadhara Rao and Singh 1999;
Thus, the liquid phase change or water vapor enhance- Tarnawski et al. 2002). Soil with higher quartz
ment induced by high temperature and the heat flux of content has larger thermal conductivity values
fluid convection due to high temperature gradient are (Tarnawski et al. 2009).

123
Geotech Geol Eng

Table 1 Thermal properties of common components in soil

Material Density Heat capacity Thermal conductivity Thermal diffusivity
(kg/m3) (kJ/kg K) (W/m K) (m2/s) 9 10-7

Air (10 °C) 1.25 1.000 0.0026 0.21

Water (25 °C) 999.87 4.200 0.56 1.43
Water vapor (1 atm, 400 K) – 1.901 0.016 233.8
Ice (0 °C) 917 2.040 2.25 12
Quartz 2,660 0.733 8.40 43.08
Granite 2,750 0.890 1.70–4.00 *12
Gypsum 1,000 1.090 0.51 4.7
Limestone 2,300 0.900 1.26–1.33 *5
Marble 2,600 0.810 2.80 13
Mica 2,883 0.880 0.75 2.956
Clay 1,450 0.880 1.28 10
Sandstone *2,270 0.710 1.60–2.10 10–13
Source Bejan and Kraus (2003)

Particle size and gradation Heat flux between Porosity The lower the void ratio, the higher the
particles is proportional to the radius of the particles. thermal conductivity (Brandon and Mitchell 1989;
Larger particles and fewer contacts in a given volume Yun and Santamarina 2008).
result in higher thermal conductivity (Aduda 1996; Cementation Cement and colloidal precipitation at
Gangadhara Rao and Singh 1999). Well-graded soil particle contacts increases the contact area and thus
exhibits higher heat transfer as small particles fill the increase the thermal conductivity (Tarnawski et al.
interstitial pore space and increases the inter-particle 2002).
coordination (Esch 2004). Through the analysis of these factors affecting the
Packing geometry Higher inter-particle coordina- effective thermal conductivity of soils found in the
tion increases the thermal conductivity for a given literature, they can be generalized into few essential
particle size (Lambert and Fletcher 1997b; Tarnawski elements: thermal conductivity of each constituent
et al. 2002). The contact conductance is more impor- (i.e., minerals, liquid, and air), water content, soil type
tant than the radiational conductance (Lambert and (e.g., particle size/shape), and particle contacts (e.g.,
Fletcher 1997a). The thermal conduction at contacts coordination numbers), which can be affected in a
results in percolation-type conduction process (Sahimi number of macro-level manifestation such as porosity,
and Tsotsis 1997). stress level, and gradation. The key governing factors
Stress level Higher stress results in higher contact are summarized in Table 2.
radii leading to an increase in thermal conductivity
therefore thermal conductivity. Granular chains under
higher stress also gives rise to more developed heat 3 Thermal Conductivity Models
transfer paths (Vargas and McCarthy 2001).
Water content Fluid volume fraction is a dominant The accurate prediction of the thermal conductivity of
factor to the thermal conductivity in partially saturated composite materials comprises a significant portion of
soils. At residual water content region, adding small the literature about heat transfer in porous media, and a
amount of water dramatically improves the thermal large number of effective thermal conductivity models
conduction. The increase of thermal conductivity with have been proposed. New models for thermal con-
increase of water content in unsaturated soils suggests ductivity of soils are emerging, suggesting that, to-
the important role of the pore fluid conduction (Singh date, a unified model or prediction procedure has not
and Devid 2000). been found with universal applicability. In this

123
 Geotech Geol Eng

Table 2 Controlling factors in thermal conductivity variation of unsaturated soils

Key governing Features
factors

Constituent Thermal conductivity of solid minerals dominates the bulk property

Soil type Different type of soil has different soil minerals, particle sizes and shapes
Water content The volume fraction of water phase determines the thermal conductivity of partially saturated soils
Particle contact The coordination number affected by stress level, packing density, gradation et al. defines the thermal
conduction skeleton

Fig. 1 Schematic
illustration of the basic
series and parallel models of
thermal conductivity:
a series model, and 3 1 3
⎛ φi ⎞
b parallel model keff = (φi ⋅ ki ) keff
= ⎜ ⎟
i =1 ⎝ ki ⎠
i =1

section, several existing models have been reviewed and parallel model. The series model imposes a
and categorized into the following three groups based constant heat flux through each serially connected
upon their principles: component (Fig. 1a) so that each component develops
Mixing models This type of model conceptualizes different temperature gradients depending on its own
the multi-phase soil system as a certain combination of thermal conductivity. In this case, the resistivity
series and parallel solid, air and/or water blocks in the (inversion of conductivity) of the bulk material is the
cubic cell or representative elementary volume arithmetic average of the resistivity of each compo-
(REV); and the effective thermal conductivity of the nent weighted by their volume fractions (harmonic
bulk medium is calculated by mixing those blocks. average for conductivity). The parallel model, on the
Empirical models This group of models builds the other hand, imposes the identical temperature gradient
relationship between relative thermal conductivity and to the individual phases or elements (Fig. 1b) so that
degree of saturation or water content, by normalizing each phase has the same temperature difference, but
the effective thermal conductivity or Kersten number conducts different heat flow depending on the thermal
over the difference between saturated state thermal conductivity of each component. In this case, the
conductivity ksat and dry state thermal conductivity thermal conductivity is the arithmetic mean of the
kdry. thermal conductivity of each component. These two
Mathematic models These models were adopted models are also referred to as upper and lower bounds
from predictive models of other physical properties, (or Wiener bounds). They are the widest bounds with
such as dielectric permittivity, magnetic permeability, least constraint since no soil structure or fabric is
electrical conductivity, and hydraulic conductivity; considered. The larger the difference between the
which are calculated by certain mathematical algo- thermal conductivity values of each component, the
rithm given the thermal conductivity of each compo- wider the bandwidth of the bounds. Some other mixing
nent and their volume fractions. laws also generate ‘‘averaged’’ effective thermal
conductivity, such as the geometric mean law (here-
3.1 Mixing Models after Geo_Mean) and the quadratic parallel law (see
Table 3).
The most common models are based on classical Another group of mixing models can be derived
mixing laws (arithmetic and harmonic) of series model based upon a combination of series and parallel

123
Geotech Geol Eng

Table 3 Selected mixing models and mathematic models for component, the key parameters a, b, c can be
effective thermal conductivity of soils determined by porosity, and formation factor bor-
Models Equations rowed from Archie’s law. This model only works for
P3 two-phase systems such as saturated soils, and deter-
Series 1
¼ i¼1 /i  k1i mination of those parameters are difficult and by
keff
P3
Parallel keff ¼ i¼1 / i  ki empirical fitting.
Geometric Q
3
/
McGaw’s (1969) model neglected the heat con-
mean keff ¼ ki i duction through particle contacts and considered the
i¼1
P pffiffiffiffi2 heat flow mainly across solids with intervening fluid
Quadratic 3
keff ¼ i¼1 /i  ki
parallel and passing entirely within the fluid (see Fig. 2c). This
Effective P3  ki keff  cubic cell can be expanded by adding one block in
i /i ki þ2keff ¼ 0
medium parallel for heat conducted in air. McGaw defined an
P 1
Self-consistent 3 /i interfacial efficiency factor e (at solid/fluid interfaces),
keff ¼ 13 i¼1 ki þ2keff
that turns to be close to 1- by assuming there is little
De Vries keff ¼ /w k/w þFs /s ks þFa /a ka
, Fi and gi are weight temperature gradient within the intervening fluid, and
w þFs /s þFa /a
(1963) and shape factors
P3    the volume of this interfacial fluid nc is about 0.03 for
Hashin– i¼2 /i 1 þ cLi high saturation sand. To use this model, some
L
Strikmann kHS ¼ k1 þ 3k1 P3 ;
/1 þ i¼2 ð/i cLi =ð1 þ cLi ÞÞ uncertain assumptions have to be made regarding the
cLi ¼ 3k1 =ðki  k1 Þ values of e and nc, when the soil is under low degrees
P2   
U i¼1 /i 1 þ cU
i
of saturation.
kHS ¼ k3 þ 3k3 P2 ;
/3 þ i¼1 ð/i cU U
i =ð1 þ ci ÞÞ
Gori and Corasaniti’s (2002) model used the cubic
cU
i ¼ 3k3 =ðki  k3 Þ
mixing model for thermal conductivity with consid-
erations of different regimes of water content for
Key factors: /i—volume fraction of each phase, i can be a (air)
three-phase soil in unsaturated conditions (see
w (water) s (solid), [–]; ki—thermal conductivity of each phase,
(ka = 0.56, kw = 0.026), (W/mK) Fig. 2d). The entire range of water content for a given
soil is divided by the water content at the field capacity
of the soil hf (when water has drained out of the larger
models. Mickley (1951) model involves a unit cubic pores but the small pores remain filled with water),
cell (see Fig. 2a) that consists of heat conduction water content at the permanent wilting point hp (water
through a (1 - a)2 solid block, heat conduction content at *1.5 kPa of suction pressure), and water
through a c2 air block, and heat conduction through content of adsorbed water films hc (water film around
a series of solid, water film, and air, assuming a water the solid particles without connecting others, a fraction
film having a thickness of b = a - c. This model of hp). The model involves arrangement of the solid
considers unsaturated soil conditions by introducing phase in the center of the gas cubic, with water films/
the parameters a, b, c, which represent the proportions bridges growing around the particle depending on the
of solid, air, and water phases that can be determined water content regime. The effective thermal conduc-
for given porosity n and degree of saturation S. How- tivities of soils in each regime are defined separately
ever, this method does not hold for very porous or dry by the mixing of air/water/solid phase. However, the
soil, due to the poor grain-to-grain contacts when the dividing points were determined empirically for
soil is loose or few water bridges Table 4. different types of soil.
Woodside and Messmer’s (1961) model employed
a three-component model that was originally used for 3.2 Empirical Models
determining electrical conductivity. The model visu-
alizes three parallel heat flow paths in an idealized unit A number of empirical relationships between effective
cube of soil as shown in Fig. 2b. It consists of a path of thermal conductivity and degree of saturation, poros-
width a through fluid and solid in series, a path of ity and soil types have been established by researchers.
width b through continuous solid material, and a path Kersten (Kersten 1949) performed an extensive series
of width c through continuous pore fluid. Given the of tests on various types of clays, silts and sands, and
thermal conductivity and volume fractions of each proposed empirical equations of thermal conductivity

123
 Geotech Geol Eng

(a)
keff = ka ⋅ c 2 + ks ⋅ (1 − a ) + kw ⋅ b 2
2

2kw ka cb 2k k (1 − a ) b 2ks kw ka (1 − a ) c
+ + s w +
kw c + ka (1 − c ) ks a + ka (1 − a ) ks kw c + ks ka b + kw ka (1 − a )

⎧ 3a 2 + 2a 3 = n
⎪
where: ⎨3c 2 − 2c3 = (1 − S ) n
⎪ c = a −b
⎩
(b)

k w ks a
keff = ks ⋅ b + k w ⋅ c +
kw d + ks (1 − d )
⎧ a +b + c =1
where: ⎨
⎩b + ad = 1 − n

(c)
ε ks (1 − n + nc )
keff = kw ⋅ ( nS − nc ) + (1 − n + nc ) + ( n − nS ) ka
ks
(1 − n ) + nc
kw
ΔTi
where: ε = 1 −
ΔT

(d)

Fig. 2 Schematic illustrations of the mixing models of thermal c McGaw’s (1969) model for unsaturated soils, and d Gori and
conductivity: a Mickley’s (1951) model for unsaturated soils, Corasaniti’s (2002) model for unsaturated soils
b Woodside and Messmer’s (1961) model for saturated soils,

based on water content and dry bulk density for silt– number and degree of saturation S. Simple first-order
clay and sandy soil with separate equations (see logarithmic functions of S for Ke were used to describe
Table 3). This empirical model leads to a discrepancy such relationships, where parameters in these equa-
of *25 % or higher for high silt–clay content soils tions were obtained by fitting the experimental results
and is only valid for the range of water content over of different types of soils. To project the Kersten
which the experimental data is available. number to the effective thermal conductivity, the
Johansen introduced a normalized thermal conduc- bounds at dry and saturated condition were calculated
tivity called Kersten number Ke, given by Ke = (k - by thermal conductivity of single phase (solid, water,
kdry)/(ksat - kdry) (Johansen 1975). A model was air) and other soil properties, such as dry bulk density,
developed by the relationships between the Kersten porosity, and quartz content. Johansen’s model

123
Geotech Geol Eng

Key parameters: S—degree of saturation, S = /w/(1 - /s), [–]; h—volumetric water content, h = ðn  SÞ, [–]; n—porosity, n = (1 - /s), [–]; cdry—dry bulk density,
provided a method to estimate the effective thermal

Sandy soil: keff ¼ 0:1442ðlog h  0:2Þ  100:6243cdry

cdry = Gs(1 - n), (kg/m3); kq—thermal conductivity of quartz, (W/mK); ko—thermal conductivity of organic materials, (W/mK); ke—Kersten number, ke = (keff - kdry)/(ksat -
Silt–clay: keff ¼ 0:1442ðlog h  0:2Þ  100:6243cdry
conductivity by interpolating between the dry and the
saturated values of the thermal conductivity. Yet,
issues come up when it applies to low quartz-content
soils.
Coarse sand: ke ¼ 0:7 log S þ 1:0

keff ¼ ks1n kwn n  ½ð1  bÞS þ bcn

Inspired by the Johansen model, some other
empirical models were derived based on the relative
Fine soil: ke ¼ log S þ 1:0 thermal conductivity described using the Kersten

ke ¼ exp½að1  Sa1:33 Þ
number Ke. Cote and Konrad modified the logarithmic
equation into a hyperbolic equation for Ke - S with a
fitted material parameter j, and redefined a new
ke ¼ 1þðj1ÞS

equation of determining dry thermal conductivity with

jS
Equations

two empirical parameters v and g (Côté and Konrad

2005). The Cote and Konrad 2002 model is applicable
to more soil types and generates more accurate
estimation of thermal conductivity for the effective
Others

thermal conductivity than the Johansen 1975 model.

b, c
v
a

Other derivative models modified based upon the

Johansen (1975) model are Lu et al. (2007) model,
2700þ0:947cdry
0:135cdry þ64:7

gn

which used a modified exponential equation of

¼ v  10

saturation S with a fitted soil-type parameter a, and

Chen (2008) model which adapted a modified power
¼

equation with two empirical parameters b and c to

kw kdry

kw kdry

describe the effective thermal conductivity behavior

ð1nÞ n

ð1nÞ n

with saturation S (Chen 2008; Lu et al. 2007). These

ke, kdry, ksat

models better consider the tail behavior of thermal

ks

ks

conductivity for fine-textured soil at low water con-

ksat ¼

ksat ¼
Table 4 Selected empirical models for effective thermal conductivity of soils

tent; however, lose applicability to other soil types.

–

Another saturation-based model is the linear interpo-

Empirical parameters: a, v, b, c—fitted parameters for different materials

lation of thermal conductivities between dry and

cdry

–
–

saturated states (e.g., Gens et al. 2009).

–

–
–
–
n

3.3 Mathematical Models

S (h)

This group of models usually is analog of other

–

–
–
–

physical properties of mixtures, such as electrical

conductivity, magnetic permittivity, elastic modulus
ð1qÞ

ð1qÞ

ð1qÞ

(see Table 3). The Hashin–Shtrikman’s model, here-

Parameters

kqq ko

kqq ko
ks ¼ kqq ko

inafter HS (Hashin and Shtrikman 1962, 1963),

ks ¼

ks ¼

employed variational principles and established the

Ki

effective conductivity bounds that were first used for

magnetic permeability and were found to be the best or
Côté and Konrad (2005)

narrowest possible bounds for composite materials,

whose bulk properties could be derived from the
Johansen (1975)

property of the constituents and their volume fractions

Lu et al. (2007)
Kersten (1949)

(Carson et al. 2005). The assumptions of HS model are

Chen (2008)

that the composite materials are macroscopically

kdry), [–]
Models

homogeneous, isotropic, multi-phase, and it considers

that the spheroidal intrusions are dispersed in a

123
 Geotech Geol Eng

different matrix. It is referred to as ‘‘external porosity’’ phase into a simple combination of series and parallel
material that the low conductivity spherical phase elements in a REV to substitute the apparent effective
(e.g., air) dispersed in the high conductivity phase property of the bulk material. Although some more
matrix (e.g., grains), restrained by the upper bound sophisticated models derived based on those mixing
such as foams or sponges. The opposite case of high principle are trying to describe the effect of pore-water
conductivity spherical phase (e.g., grains) dispersed in distribution (e.g., Gori and Corasaniti (2002) model) or
low conductivity matrix (e.g., air or water) is the simulate the soil microstructure by fractal model (e.g.,
‘‘internal porosity’’ material such as soils, restrained Lehmann et al. 2003 model), they are basically lump
by the lower bound (see schematic illustration in effective bulk properties without a realistic physical
Fig. 3a). The HS upper and lower (HS_U, HS_L) mechanism. The empirical models interpolate the
bounds always lie within the parallel/series bounds effective thermal conductivity between the minimum
discussed above, regardless of the component volume value (dry thermal conductivity) and the maximum
fractions or thermal conductivities. value (saturated thermal conductivity) by using differ-
The effective medium theory (EMT), as one kind of ent types of functions with empirical fitted parameter to
the self-consistent method (SCM), has been proposed apply different types of soil, whereas each model is only
to distinguish the thermal conductivity region bounded valid for a small group of soil type, and those parameters
by HS bounds into internal porosity and external lack clear physical meanings, plus the determination of
porosity (Carson et al. 2005). The EMT model assumes dry and saturated thermal conductivity were also not
that the bulk property of a composite material as a definite. The mathematical models approximate the
result of the interaction of each phase is accounted for effective thermal conductivity only by given property of
by imaging each phase to be an inclusion embedded in each component and their volume fractions, by homog-
a homogeneous medium that has the overall property enization assumptions and analog of phase intrusion of a
of the composite. It is equivalent to the composite that matrix in an electrical field or elastic energy conserva-
has no continuous or dispersed phases with different tion of composite material under a field constrain,
conductivity; instead, the different phases are ran- however, all of them have no consideration of the soil
domly distributed with the medium (see schematic microstructure, solid particle geometry, water-phase
illustration in Fig. 3b). distribution, and pore-size gradation.
In summary, the mixing models are trying to simplify
the soil microstructure and the arrangement of each
4 Model Assessment by Experimental Data

In this section, the thermal conductivities of collected

experimental data of different types of soil are
compared with the prediction of the abovementioned
models. Cross validation was used to evaluate each
model. The properties of three types of soil used in the
Fig. 3 Schematic illustrations of the mathematical models of model predictions and assessments are listed in
thermal conductivity: a Hashin–Shtrikman or HS model, and Table 5. The normalized root-mean-square-error
b self-consistent model (NRMSE) and the coefficient of variation of the root

Table 5 Physical properties used in model comparisons

Experimental data Parameters
ks (W/mK) Gs (unitless) n (unitless) cdry (kg/m3) kdry (W/mK)a ksat (W/mK)

Sands Accusand 40/50 4.75 2.65 0.37 1.696 0.07/0.28 2.27

Silt Bonny silt 2.5 2.53 0.39 1.25 0.06/0.26 1.42
Clay Tarnawski et al. (2002) 2.5 2.65 0.5 1.325 0.06/0.19 1.18
a
The values are calculated by equation in Johansen (1975) and Côté and Konrad (2005) respectively

123
Geotech Geol Eng

Table 6 Cross validation of the experimental data and model predictions

Cross validation Mixing models Mathematical models Empirical models
Parallel Series Geo_mean HS_L HS_U SCM De Vries Johansen Côté and Lu et al.
(1952) (1975) Konrad (2005) (2007)

Sand
NRMSE 0.631 0.819 0.395 0.747 0.456 0.333 0.253 0.374 0.282 0.330
CV(RMSE) 0.745 0.966 0.466 0.881 0.538 0.393 0.299 0.441 0.333 0.389
Silt
NRMSE 1.019 0.788 0.219 0.628 0.769 0.576 0.295 0.224 0.145 0.310
CV(RMSE) 1.097 0.848 0.236 0.676 0.828 0.620 0.317 0.241 0.156 0.334
Clay
NRMSE 0.839 0.612 0.175 0.462 0.635 0.437 0.311 0.297 0.355 0.255
CV(RMSE) 0.997 0.728 0.208 0.549 0.755 0.520 0.370 0.353 0.422 0.303

mean square error CV(RMSE) defined for the three

sets of experimental data and the selected model
predictions are summarized in Table 6. These values
are defined as follows:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P 
kexp  kmod
RMSE ¼ ð4Þ
n

RMSE
NRMSE ¼ max  k min
ð5Þ
kexp exp

RMSE
CV ðRMSEÞ ¼ ð6Þ
kexp
The smaller the NRMSE number, the less the model
prediction deviates from the measured data points.
A comparison of a set of measured thermal
conductivity values for quartz sand and the calculated
thermal conductivity by different models is shown in
Fig. 4 for the conditions listed in Table 5. In Fig. 4a, it
can be clearly seen that the parallel/series models set
the upper and lower limits of the possible thermal
conductivity predictions. These two limits have the
largest bandwidth, due to the large difference in the
thermal conductivity values between the solid mineral
and the air. The Geometric Mean model has an upward
concave trend as the saturation increases, across the Fig. 4 Comparison of thermal conductivity model simulation
convex experimental data points. The HS upper and and the experimental data of sands: a effective thermal
lower bounds lie within the parallel and series bounds, conductivity with degree of saturation (group I uses solid lines,
but still have large width and fail to capture the group II dot-dash lines, and group III dot lines), and b 1:1
comparison space for predicted and measured values (group I
thermal conductivity variation pattern. All of the uses symbols minus, plus, mulitiplication; group II hollow
experimental data points fall into the external porosity symbols square, dimond, triangle, circle; group III solid-filled
region (lower than the SCM model, short dash–dot symbols square, triangle, circle)

123
 Geotech Geol Eng

Fig. 6 Comparison of thermal conductivity model simulation

Fig. 5 Comparison of thermal conductivity model simulation and the experimental data of clay: a effective thermal
and the experimental data of silt: a effective thermal conduc- conductivity with degree of saturation (group I uses solid lines,
tivity with degree of saturation (group I uses solid lines, group II group II dot-dash lines, and group III dot lines), and b 1:1
dot-dash lines, and group III dot lines), and b 1:1 comparison comparison space for predicted and measured values (group I
space for predicted and measured values (group I uses symbols uses symbols minus, plus, mulitiplication; group II hollow
minus, plus, mulitiplication; group II hollow symbols square, symbols square, dimond, triangle, circle; group III solid-filled
dimond, triangle, circle; group III solid-filled symbols square, symbols square, triangle, circle)
triangle, circle)

line). The SCM model is close to the experiment material. Figure 4b shows the 1:1 line for comparison
results only at high degrees of saturation. The De Vries of the measured and calculated results. None of the
(1963) model underestimates the thermal conductivity models have data points falling around the diagonal
at most of the range of saturation (De Vries 1963). All line throughout the entire range of saturation. The
the empirical models have similar calculation results NRMSEs of all models are large (0.253–0.819),
of the thermal conductivity at different saturations. indicating that very large discrepancies exist between
The relative thermal conductivity relations defined by the experimental data and model predictions.
the models of Johansen (1975), Côté and Konrad Although, the empirical models have lower NRMSE
(2005) and Lu et al. (2007) use logarithmic, exponen- values than the other models, they still largely
tial and power functions, respectively, and have a underestimate the thermal conductivity.
similar trend with the measured data points. The Figure 5 delineates the comparison of experimental
functions describing those relations generate smooth measurements of sandy silt and model calculations.
curves and therefore cannot reflect the characteristic Evidently, the parallel/series and HS upper/lower
inflection points, which represent the critical satura- bounds set the upper and lower limits of the thermal
tion values dividing the pore-water distribution conductivity values. Geometric mean model has the
regimes, on the thermal conductivity curve for sand right trend but concaves to an opposite direction to the

123
Geotech Geol Eng

experimental data. SCM model has ‘‘quasi-linear’’ non-uniform. SCM further lowers the upper bound by
projections and higher predictions over all the satura- dividing the region bounded by HS model into
tion range. De Vries (1963) model has close results at ‘‘internal’’ and ‘‘external’’ porosity parts, making
mediate saturation range 0.2–0.6. Empirical models SCM model better in upper bound predictions. The
are quite fit with the experimental data points. thermal conductivities of soil at partially saturated
Nevertheless, only Lu et al.’s (2007) model captures conditions are always lower than SCM line where the
the characteristic ‘‘flat-tail’’ behavior of thermal ‘‘external porosity’’ region is resulted from the with
conductivity of silt at low degrees of saturation with high-thermal-conductivity soil particles ‘‘dispersed’’
a slightly shift. Also Johansen’s and Lu et al.’s models in relatively low-thermal-conductivity air or water.
underestimate the thermal conductivity at lower This is not physically realistic in unsaturated soil. The
saturation and overestimate at higher saturation. Geometric Mean model averages the thermal conduc-
Cote’s model has the best fit with experimental data tivity of each phase in the unsaturated soil mixture but
but fails to simulate the stasis behavior of thermal overestimates the contribution of pore-water at low
conductivity of silt at low degrees of saturation. degree of saturation and underestimate it at high
Figure 5b shows the 1:1 diagonal comparison, and degree of saturation. The actual thermal conductivity
Côté and Konrad (2005) model has the lowest NRMSE for unsaturated soils evolves from the HS lower bound
value of 0.145. at dry and approaches to the SCM line at full
Figure 6 shows the comparison of experimental saturation. As mentioned, the averaging and homog-
data of clay soil and the model calculations. Parallel/ enization techniques do not consider the air–water
Series model and the HS model again set the upper and interface and pore-water distribution in the soil at
lower bounds of the thermal conductivity for the different stages of saturation. They do incorporate the
multi-phase mixture. Geometric mean always cross governing factors of constituent, and water content
the experimental curve and has the different shape of (given the thermal conductivity of each phase and their
the curve. SCM model put all experimental data points volume fraction), but miss the factors of soil type and
into the ‘‘external porosity’’ zone between the SCM particle contacts (i.e., the influences of particle
and HS_L line. Other mathematical models have geometry, mineral type, and particle connectivity),
higher calculation result than the experimental results. and different water retention mechanisms for different
All empirical models have monotonically increases of saturations.
thermal conductivity simulation as the saturation The empirical models use certain mathematical
increases, which are all far away from the true functions to fit the experimental results and then
sigmoidal behavior for clay. Figure 6b shows the 1:1 determine the parameters in these functions to match
diagonal comparison, with the NRMSE values ranging with certain type of soil. The difficulty of this group to
from 0.175 to 0.839. apply all soil types is the uncertainty in calculating
By the above comparisons, the parallel/series thermal conductivity of the solid minerals, which
bounds are much higher/lower than the real thermal makes it harder to determine two end members, the
conductivity of the sand measured throughout the thermal conductivity of dry and saturated states.
saturation range. However, they set theoretical limits Additionally, the smooth curves generated by loga-
for all model predictions. The reason for that lies in the rithmic, exponential or power equations also cannot
fact that the parallel/series models assume each of the represent the different water retention regimes
components are stacked in layers to form the multi- observed in different types of soil. The empirical
phase mixture. Those two represent the extreme cases models work better for silt and clay than sand, but are
of the heterogeneity of material and geometry. HS still far from accurate when soil has rich clay-content
bounds have an additional constraint by assuming the at low degrees of saturation. The empirical models
second or other phases are uniformly dispersed in a integrate the key factors of constituent in terms of the
continuous matrix. This homogenization technique thermal conductivities at dry state kdry and at full
generates more realistic upper and lower limits and saturated state ksat, and consider the factor of water
thus narrows the bandwidth between two limits for the content, but still omit the factors of soil type and
thermal conductivity of multiphase composites. How- particle contact, which depends on the micro-structure
ever, water distribution in multiphase soil is generally of the material.

123
 Geotech Geol Eng

Fig. 7 Conceptual model of thermal conductivity: a thermal conductivity and soil–water retention, and b regimes of pore-water
distribution

5 A Unified Conceptual Model increase until the degree of saturation increases up to a

certain threshold. The higher fines content of a soil, the
The change in thermal conductivity of soils with longer the flat-tail development of thermal conductiv-
increasing degree of saturation differs remarkably for ity occurs at low degrees of saturation (e.g., clay has
different soil types. For sandy soils, the thermal longer stasis development of thermal conductivity at
conductivity increases immediately once the soil low degrees of saturation, and therefore larger value of
experiences a small increase in degree of saturation, such threshold of saturation than silt). As the satura-
with a sharp jump to almost 70 % of the heat transfer tion continues increasing, the thermal conductivity
capability at full saturation during wetting from dry gradually increases to gain most of the heat transfer
conditions to approximately 5–10 % degree of satu- capability in a short range of saturation (e.g., 5–20 %
ration. The increase in thermal conductivity becomes for silt, and 35–55 % for clay in this case). Next, the
inconspicuous as the soil approaches fully saturated thermal conductivity slowly increases to the maxi-
conditions. Inflection points are clearly observed for mum value and then keeps constant, as the saturation
those separate regions at different critical degrees of increases. Generally the thermal conductivity of silt
saturation, indicating transition from one heat transfer and clay is lower than that of the sand over the entire
mechanism to another. range of saturation, due to the lower thermal property
For silt and clay soils, it can be observed that the of mineral constituents consist of fine soil than sands.
thermal conductivity evolution with changing degree Based on the afore-mentioned analysis and obser-
of saturation is much smoother than sand, without vation, a conceptual model unifying thermal conduc-
sharp inflection points. It can also be observed that the tivity behavior and soil water retention mechanisms is
thermal conductivity values of silt or clay do not proposed and shown in Fig. 7. The full range of

123
Geotech Geol Eng

thermal conductivity variation with saturation can be low saturation region (e.g., S = 0.05–0.1). However,
distinguished and conceptualized in four regimes: silty and clayey soil can have considerable amount of
hydration, pendular, funicular, and capillary. The hydration water, which prevails the low saturation
range and boundary of each regime depends on the region and gradually transits to the pendular regime.
soil and liquid type and micro-structures (i.e., particle The hydration water makes the thermal conductivity
geometry, particle/pore size distribution, pore-water almost constant at the early stage of wetting, and the
arrangement, and interfacial properties), which can be overlap between hydration and pendular regime
identified as the key governing factors shown in makes the rapid smooth increase of thermal conduc-
Table 2. These factors have been largely ignored in all tivity of silt or clay as the water content increase but
the existing models. More importantly, these factors not as steep as that of sand. The significant increase in
can also be characterized by the soil–water retention water meniscus at the particle contacts is the physical
curve (SWRC). At thermodynamic or multiphase reason why thermal conductivity in all types of soil
equilibrium for a given saturation, matric suction increases rapidly within the pendular regime. The
value indicates the information of certain pore size pendular regime can be S = 0.05–0.15 for silt or
that holds certain amount of pore water of that soil. 0.35–0.55 for clay. Of all the previous models, Only
The characteristic changing of slopes and turning Cote’s equation predicts closely to the measured data
features on the SWRC curves reflect the effects of soil in this regime for silt. All empirical models fail to
and liquid type and pore size distribution. Therefore, capture the rapid change of the thermal conductivity of
SWRC is a promising tool to unify all significant sand in this regime, and these models also cannot
governing factors of the thermal conductivity of accurately predict thermal conductivity variations in
unsaturated soils for the proposed conceptual model. clayey soil.
As illustrated in Fig. 7, from dry state to the first As the saturation continues to increase, the liquid
critical degree of saturation defines the hydration state bridges grow and begin to merge with adjacent ones
(i.e., regime 1), where only hydration water formed and start to form a thick connected water-film around
with the interaction of clay minerals. In this stage, the soil particles, fill up the void space of pores. This
water molecules are absorbed into mineral complex, regime carries on the buildup of the pore-water
which do not change the soil particle network or network, and this is the physical reason why the
connectivity, thus the thermal properties of each connectivity of heat transfer paths is further enhanced,
constituent in soil. This is the physical reason why in leading to a gradual increase of thermal conductivity
rich fine-content soil like silt and clay at the low in funicular regime 3 (Fig. 7). Because of the high
degree of saturation, the thermal conductivity of bulk thermal conductivity of solid minerals and the pore-
soil barely changes (e.g., S from 0 to 0.05 for silt or water network, the heat flow goes in preference of the
0.35 for clay). The higher the fines content, the wider solid skeleton connected by pore-water at particle
the range of this regime, and the higher value of this contacts. Hence, further increase of water content does
first critical saturation value. In the hydration regime, not remarkably improve the thermal conductivity of
although pore water has been involved, the soil system the system as much as that in pendular regime. Here,
behaves still like a two-phase mixture (solid/air), the thermal conductivity of soil gradually increases to
where the thermal conductivity follows the path close approach the maximum value (e.g., S = 0.1–0.8 for
to the lower bounds (series/HS lower). soil, 0.15–0.85 for silt and 0.55–0.85 for clay). The
As the saturation increases, pore water starts to three empirical models have good fit in this section
form thin film around the soil particle and build with silt, but underestimate the values of sand and
individual water bridges at the particle contacts, which overestimate the values of clay.
is defined as pendular regime 2. The water film/bridge At even higher saturation, the pore-water occupies
overcomes the huge contact resistance in the dry soil- most of the voids instead of air, which does not further
air–solid contact conduction and significantly increase the thermal conductivity of the soil mixture in
improves the connectivity of thermal transfer paths, capillary regime 4 (Fig. 7). In this regime, the increase
which contributes the rapid increase of the thermal of the pore water either does not further alter the
conductivity. Sandy soil has very limited hydration preference of heat flow through the soil skeleton or
stage; therefore the pendular regime dominates the improve the connectivity of heat transfer pathway. The

123
 Geotech Geol Eng

range of capillary regime for sand is around 0.8–1, and either work only at portions of saturation range or apply
0.85–1 for silt and clay. From mediate to high degree to certain soil type. Most importantly, these models
of saturation, all models are converging to the upper lack the consideration of some key governing factors
bound (SCM line). Nevertheless for the empirical such as soil and liquid type and pore size distribution. It
models, which are better than others, the performance is found that the pore-scale water distribution based on
is limited by the accurate calculation the saturated thermodynamic equilibrium conditions among air–
thermal conductivity, which is estimated by a geo- water–solid interfaces during different saturation
metric mean of solid and liquid phase. states have not been reflected in all the existing models.
As shown, by linking the water retention mecha- A conceptual model is proposed based on the
nisms and thermal conductivity variation behavior, analysis of heat transfer mechanisms and the discus-
and by comparing the previous models with the sion about missing factors from existing models in
experimental data for different types of soil, the large terms of the effect of pore water distribution at
discrepancies predicted by the previous models lay in different water retention regimes. These water reten-
the overlook of the roles of different water retention tion regimes are governed by the SWRC and can be
regimes in thermal conductivity variation. None of the used to fully reconcile variations of thermal conduc-
models explicitly used the governing factors of soil tivity with degree of saturation for all types of soil.
and liquid types and pore size distribution that Quantitative linkages between the pore size distribu-
characterize the soil water retention. The existing tion, soil–water interaction mechanisms, and the bulk
models are all pore-structure independent. In two- thermal conductivity variation behavior can be estab-
phase systems, the previous theoretical models are lished in light of the SWRC.
close to experiment results for certain type of soil.
However, when air and water are both involved, the Acknowledgments The funding for this research is provided
by a grant from National Science Foundation (NSF-CMMI-
complexity of pore-water network and connections is
1230544) to JSM and NL.
increased due to the air–water interfaces. How water
phase distributed in the pore space at different degrees
of saturation defines the connectivity of the pore water
References
in the soil skeleton and therefore determines the
thermal behavior of unsaturated soils. Aduda BO (1996) Effective thermal conductivity of loose par-
ticulate systems. J Mater Sci 31:6441–6448
Bejan A, Kraus AD (2003) Heat transfer handbook. Wiley, New
York
6 Conclusions Brandl H (2006) Energy foundations and other thermo-active
ground structures. Geotechnique 56:81–122
This paper reviews the existing models of thermal Brandon TL, Mitchell JK (1989) Factors influencing thermal
conductivity for unsaturated soil, assess these models’ resistivity of sands. J Geotech Eng 115(12):1683–1698
Carslaw HS, Jaeger JC (1959) Conduction of heat in solids, 2nd
predictability, and thereby identify the controlling
edn. Oxford, London
factors for thermal conduction in porous media. The Carson JK, Lovatt SJ, Tanner DJ, Cleland AC (2005) Thermal
existing thermal conductivity models are assessed and conductivity bounds for isotropic, porous materials. Int J
categorized into three groups based on their assump- Heat Mass Transf 48:2150–2158
Chen S (2008) Thermal conductivity of sands. Heat Mass Transf
tions and principles. Mixing models simulate the 44:1241–1246
thermal conductivity of mixtures by a combination of Cortes DD, Martin AI, Yun TS, Francisca FM, Santamarina JC,
simple parallel or series elements consisting of the Ruppel C (2009) Thermal conductivity of hydrate-bearing
composite. Mathematical models adopt the homoge- sediments. J Geophys Res 114:B11103
Côté J, Konrad J-M (2005) A generalized thermal conductivity
nization approaches to calculate the thermal conduc-
model for soils and construction materials. Can Geotech J
tivity of a uniform composite. Empirical models use 42:443–458
different functions to fit the experimental measure- De Vries DA (1952) The thermal conductivity of soil. Med-
ments and their parameters to apply to certain soil edelingen van de Landbouwhogeschool te Wageningen 52
(1):1–73
types. All model predictions are compared with
De Vries DA (1963) Thermal properties of soils. In: Van Wijk
experimental data of three types of unsaturated soil: WR (ed) Physics of plant environment. Wiley, New York,
sand, silt and clay. It is found that these models are pp 210–235

123
Geotech Geol Eng

Ebigbo A (2005) Thermal effects of carbon dioxide sequestra- complex heterogeneous materials system: application on
tion in the subsurface, Master’s thesis, Institut für Was- nuclear waste forms. Metall Mater Trans A 44(1):1–9
serbau, Universität Stuttgart Lu S, Ren T, Gong Y, Horton R (2007) An improved model for
Esch DC (2004) Thermal analysis, construction, and monitoring predicting soil thermal conductivity from water content at
methods for frozen ground. TCCRE Monographs. ASCE. room temperature. Soil Sci Soc Am J 71(1):8–14
Reston, VA, p 498 McGaw R (1969) Heat conduction in saturated granular mate-
Farouki OT (1981) Thermal properties of soils. Cold Regions rials. Highway Research Board Special Report
Science and Engineering, CRREL Monograph 81–1, 136 Mickley A (1951) The thermal conductivity of moist soil. Am
Gangadhara Rao M, Singh DN (1999) A generalized relation- Inst Electr Eng Trans 70:1789–1797
ship to estimate thermal resistivity of soils. Can Geotech J Mitchell JK, Soga K (2005) Fundamentals of soil behavior.
36:767–773 Wiley, New York
Gens A, Sánchez M, Guimaraes LDN, Alonso EE, Lloret A, Murashov VV, White M (2000) Thermal conductivity of crys-
Olivella S, Villar MV, Huertas F (2009) A full-scale in situ talline particulate materials. J Mater Sci 35:649–653
heating test for high-level nuclear waste disposal: observa- Preene M, Powrie W (2009) Ground energy systems: from
tions, analysis and interpretation. Géotechnique 59:377–399 analysis to geotechnical design. Geotechnique 59:261–271
Gori F, Corasaniti S (2002) Theoretical prediction of the soil Sahimi M, Tsotsis TT (1997) Transient diffusion and conduc-
thermal conductivity at moderately high temperatures. tion in heterogeneous media: beyond the classical effec-
J Heat Transf Trans Asme 124(6):1001–1008 tive-medium approximation. Ind Eng Chem Res
Hashin Z, Shtrikman S (1962) A variational approach to the 36:3043–3052
theory of the effective magnetic permeability of multiphase Singh DN, Devid K (2000) Generalized relationships for esti-
materials. J Appl Phys 33:3125–3131 mating soil thermal resistivity. Exp Thermal Fluid Sci
Hashin Z, Shtrikman S (1963) A variational approach to the 22:133–143
theory of the elastic behaviour of multiphase materials. Smits KM, Sakaki T, Limsuwat A, Illangasekare TH (2010)
J Mech Phys Solids 11:127–140 Thermal conductivity of sands under varying moisture and
Johansen O (1975) Thermal conductivity of soils. N.H. CRREL porosity in drainage–wetting cycles. Vadose Zone J
Draft English Translation 637. Ph.D. thesis, University of 9:172–180
Trondheim, Trondheim, Norway. US Army Corps of Tarnawski VR, Leong WH, Gori F, Buchan GD, Sundberg J
Engineers, Cold Regions Research and Engineering Lab- (2002) Interparticle contact heat transfer in soil systems at
oratory, Hanover moderate temperatures. Int J Energy Res 26:1345–1358
Kersten MS (1949) Laboratory research for the determination of Tarnawski VR, Momose T, Leong WH (2009) Assessing the
the thermal properties of soils. Technical Report 23. impact of quartz content on the prediction of soil thermal
Research Laboratory Investigations, Engineering Experi- conductivity. Geotechnique 59:331–338
ment Station, University of Minnesota, Minneapolis, Minn Vargas WL, McCarthy JJ (2001) Heat conduction in granular
Lambert MA, Fletcher LS (1997a) Thermal contact conductance materials. AIChE J 47:1052–1059
of spherical rough metals. J Heat Transf Trans Asme White DE (1973) Characteristics of geothermal resources.
119(4):684–690 Geothermal Energy, Stanford University Press, Stanford,
Lambert MA, Fletcher LS (1997b) Review of models for ther- pp 69–94
mal contact conductance of metals. J Thermophys Heat Woodside W, Messmer JH (1961) Thermal conductivity of
Transf 11:129–140 porous media. I. Unconsolidated sands. J Appl Phys
Lehmann P, Stähli M, Papritz A, Gygi A, Flühler H (2003) A 32:1688–1699
fractal approach to model soil structure and to calculate Yun TS (2005) Mechanical and thermal study of hydrate bearing
thermal conductivity of soils. Transp Porous Media sediments, Ph.D. Georgia Institute of Technology, Atlanta
52:313–332 Yun TS, Santamarina JC (2008) Fundamental study of thermal
Li D, Sun X, Khaleel M (2012) Comparison of different up- conduction in dry soils. Granular Matter 10:197
scaling methods for predicting thermal conductivity of

123