GPR analysis of clayey soil behaviour in unsaturated conditions for pavement engineering

and geoscience applications

Fabio Tosti1,*, Andrea Benedetto2, Luca Bianchini Ciampoli2, Sébastien Lambot3, Claudio

Patriarca4, Evert C. Slob5

1
School of Computing and Engineering, University of West London (UWL), St Mary's Road,

Ealing, W5 5RF, London, United Kingdom

2
Department of Engineering, Roma Tre University, Via Vito Volterra 62, 00146, Rome, Italy; e-

mail: [Link]@[Link]; [Link]@uniroma3; [Link]@[Link]

3
Earth and Life Institute, Environmental Sciences Université catholique de Louvain (UCL)

Croix du Sud 2 box L7.05.02, 1348 Louvain-la-Neuve, Belgium;

[Link]@[Link]
4
Statoil ASA, Sandsliveien 90, 5254 Sandsli, Norway; clpa@[Link]
5
Delft University of Technology, Faculty of Civil Engineering and Geosciences, Section of

Applied Geophysics and Petrophysics Department of Geoscience & Engineering, Stevinweg 1,

Delft 2628 CN, The Netherlands; [Link]@[Link]

* Corresponding author: mail: [Link]@[Link]; tel: +39 0657333543

1
Abstract

Clay content is one of the primary cause of pavement damages, such as subgrade failures,

cracks and pavement rutting, thereby playing a crucial role in road safety issues as an indirect

cause of accidents. In this paper, several ground-penetrating radar (GPR) methods and analysis

techniques were used to non-destructively investigate the electromagnetic behavior of sub-

asphalt compacted clayey layers and subgrade soils in unsaturated conditions. Typical road

materials employed for load-bearing layers construction, classified as A1, A2 and A3 by the

American Association of State Highway and Transportation Officials (AASHTO) soil

classification system, were used for the laboratory tests. Clay-free and clay-rich soil samples

were manufactured and adequately compacted in electrically and hydraulically isolated

formworks. The samples were tested at different moisture conditions from dry to saturated.

Measurements were carried out for each water content using a vector network analyzer spanning

the 1–3 GHz frequency range, and a pulsed radar system with ground-coupled antennas, 500

MHz center frequency. Different theoretically-based methods were used for data processing.

Promising insights are shown to single out the influence of clay in load-bearing layers and

subgrade soils, and its impacts on their electromagnetic response at variable moisture

conditions.

2
List of keywords

clay detection; ground-penetrating radar; SFCW; soil moisture; road pavement; Rayleigh

scattering; full-wave inversion, surface reflection method, volumetric mixing formulae

3
INTRODUCTION

Clay detection in soils is a key topic of research in many fields of application such as

construction (Miqueleiz et al. 2012), pavement (Uzan 1998) and geotechnical engineering

(Abusharar and Han 2011), agriculture (Robinson and Phillips 2001), and Earth sciences in

general (Mahmoudzadeh et al. 2011). Overall, the interaction between water and clay plays a

very important role in the health conditions of pavements due to the considerable swelling

properties of clay (Wuddivira et al. 2012), since it is capable to exert significant effects on the

stability of soils behavior under loading.

The chemical and mechanical behavior of clay, has been widely investigated in the past. As a

rule of thumb, it is conventionally established to model the hydrodynamic behavior of plastic

soils during wetting-drying cycles (Dudoignon et al. 2007). Richard et al. (2001) argued that

particle arrangements along with their mineralogical nature can notably affect the properties of

clayey soils under compressive or drying stress cycles. More recently, Beroya, Aydin, and

Katzenbach 2009 have also proved that clay minerals abundance mainly governs the cyclic

behavior of silt-clay mixtures.

Several procedures have been established over the years to limit the effects of clay in soils for

construction and rehabilitation of transport infrastructures. In this regard, the use of geogrids

(Abdi Sadrnejad, and Arjomand 2009) or additives (Pakbaz and Alipour 2012) are very

common practices. When suitable strength and deformation properties of soils cannot be

ensured by excessive abundances of plastic material, extra reinforcing steel to foundations and

slabs as well as earthmoving operations for removing waste materials are instead usually carried

out (Wood, Osborne, and Forde 1995). Among the most common destructive techniques for

clay investigation, core sampling relies on an undeniable accuracy. However, coring is a local

technique and it might not be representative of large areas; in addition to that, this method is

invasive, expensive and time consuming.

4
The demand for non-destructive testing techniques (NDTs) in this field of application and

beyond is increasing nowadays. To cite a few, acoustic methods like falling weight

deflectometer (FWD) techniques (Benedetto et al. 2014a) are widely used to evaluate the

strength and deformation properties of clayey soils, as well methods relying on spectral

analyses, such as the vis-NIR spectroscopy (Viscarra Rossel et al. 2009), or infrared imaging

(Srasra et al. 1994) of soil materials. However, most of the NDTs are not time-efficient, since

they cannot be continuously implemented on roads, nor cost-effective for road inspection and

maintenance. Amongst the main electromagnetic (EM) methods fit for purposes, advanced high-

resolution radiometer (AVHRR) (Odeh and McBratney 2000) and electromagnetic induction

(EMI) (Triantafilis and Lesch 2005) can be considered as other very effective techniques,

although the scale domain of investigation or the long time required for surveying are not

suitable for pavement engineering applications.

It is therefore clear how an effective high-performance method capable to provide large-scale

reliable measurements with a high spatial resolution, is nowadays required. In line with this,

ground-penetrating radar (GPR) has proved to be one of the most powerful diagnostic non-

destructive tools that enables to collect data rapidly in the field (Saarenketo and Scullion

2000;Benedetto et al. 2012a; Tosti et al. 2014a). GPR is being increasingly employed in a range

of many application areas such as planetary exploration, cultural heritage protection, Earth

sciences, and engineering applications (Daniels 2004). This instrument allows to infer the

physical conditions of subsurface relying on the transmission/reception of short EM impulses in

a given frequency band (Van der Kruk and Slob 2004; Slob, Sato, and Olhoeft 2010;).

GPR was first used in traffic infrastructure surveys in 1970s by the Federal Highway

Administration (FHWA) for testing in tunnel investigation (Morey 1998). The main

applications in this field range from physical to geometrical inspections of pavement layers.

They broadly include the evaluation of layer thicknesses (Al-Qadi and Lahouar 2004), the

assessment of damage conditions in hot mix asphalt (HMA) layers (Scullion, Lau, and Chen

1994), load-bearing layers and subgrade soils (Saarenketo and Scullion 2000), the inspection of

5
concrete structures (Huston et al. 1999; Benedetto et al. 2012b). New frontiers on the use of

GPR in pavement engineering have been also recently tackled on the possibility to infer

mechanical properties of road pavements and materials from their EM characteristics

(Benedetto and Tosti 2013; Tosti et al. 2014b). In addition, FDTD simulation-based approaches

of the GPR signal have been implemented for analysing the GPR responses of typical scenarios

of pavement faults (Benedetto et al. 2014b).

Many studies in the literature have been devoted towards the evaluation of water content

(Robinson et al. 2008). Applications in this field can be broadly classified according to the type

of pavement and construction material, whereby different radar systems and processing

techniques can be specifically employed. Concerning subsurface moisture measurements in

subgrade soils, GPR can bridge the gap between high-resolution data (from ~10-2 m to 10-1 m)

by minor destructive techniques, such as capacitance probes (Wobshall 1978) and time domain

reflectometry (TDR) (Fellner-Feldegg 1969), and low-resolution techniques (~10 m) by remote

sensing investigations (Wagner et al. 2007), thereby ranking effectively as an intermediate-scale

effective technology (from ~10-1 m to 1 m) (Lambot et al. 2006, Minet et al. 2011). Several

approaches exist for determining the volumetric water content (VWC) θ by GPR (Huisman et

al. 2003). An estimate of the relative dielectric permittivity εr is firstly carried out, and a

petrophysical relationship is then used to convert εr into θ. Among such expressions, the

empirical equation proposed by Topp, Davis, and Annan (1980) is undoubtedly the most used

for this purpose. From a theoretical point of view, can be also related to by using

volumetric mixing formulae, which use the volume fraction and the dielectric permittivity of

each soil constituent to derive a relationship between them (Dobson et al. 1985, Roth et al.

1990).

Overall, the aforementioned methods for moisture evaluation rely on the estimates of the

relative dielectric permittivity of soils, and the use of core sampling is mostly needed for

calibrating the system. In this regard, research activities are increasingly being focused on the

use of efficient and self-consistent techniques, namely, capable to minimize or avoid the use of

6
destructive samplings for calibration. Lambot et al. (2004a) evaluated the dielectric properties

of unsaturated soils at the laboratory scale by relating the imaginary part of the dielectric

permittivity and the frequency of investigation. Moreover, a Rayleigh scattering-based method

has been used by Benedetto (2010) and Benedetto et al. (2015) for directly predicting the

volumetric water content in soils, thereby enabling to avoid the use of any calibrations of the

system.

The GPR-based detection of clay content has been mainly investigated over the years in studies

applied at the field scale (Gómez-Ortiz et al. 2010, De Benedetto et al. 2012) and, partly, at the

laboratory scale (Saarenketo 1998). More recently, advances in the research have registered

several clay-dedicated laboratory studies specifically focused on achieving high-resolution

estimates on different textured soil materials using several self-consistent processing techniques

and different radar systems (Patriarca et al. 2013, Tosti et al. 2013).

METHODOLOGY AND OBJECTIVES

In this paper, it is analyzed the ability of GPR to detect clay in different types of soil at variable

moisture contents by using several tools and signal processing techniques. Three types of soil

classified by the American Association of State Highway and Transportation Officials

(AASHTO 2011) as A1, A2, and A3 were used for laboratory testing. To single out the

electromagnetic behavior of the above materials relative to clay presence, clay-free and clay-

rich conditions amounting up to 15% by weight were manufactured. The consistency of results

was validated through permittivity-based methods, namely, the full-wave inversion method, the

surface reflection method, and the volumetric mixing formulae. In addition, a Rayleigh

scattering-based method working in the frequency domain was employed for the analyses.

The possibility to detect clay in load-bearing layers and subgrade soils for preventing structural

failures of pavements and restraining the risk of severe damage can be considered as the main

task of this paper, as part of an important countermeasure in quality control, rehabilitation, and

7
maintenance operations made in transport infrastructures for improving transport safety

conditions.

More specifically, the aforementioned processing techniques were applied on data collected

from samples of compacted loose materials, representing conditions from load-bearing layers

and subgrade soils, which were investigated using different GPR systems.

THEORETICAL BACKGROUND

Full-Wave Inversion Technique

The radar model

Lambot et al. (2004a) proposed an intrinsic far-field antenna model combined with planar

layered media Green's functions where a local homogeneous field distribution is assumed for

the backscattered field over the antenna aperture. Therefore, the antenna radiation properties can

be described by an equivalent single electric dipole approximation. In such system, the wave

propagation between the source point and the radar transmission line reference plane, as well as

the antenna-medium interactions, can be described, on the basis of the linearity of Maxwell’s

equations, by means of complex, frequency-dependent global reflection and transmission

coefficients. Equation (1) describes the relationship between the radar-measured field and the

3D layered medium Green’s function in the frequency domain (Lambot et al. 2004a):

(1)

with representing the raw radar signal as the ratio between the backscattered field

and incident field at the radar transmission line reference plane, and describing the

angular frequency. stands for the global reflection coefficient of the antenna in free

space, while = with being the global transmission coefficient for fields

incident from the radar reference plane onto the source point, and being the global

transmission coefficient for fields incident from the field point onto the radar reference plane.

stands for the global reflection coefficient for the field incident from the layered medium

8
onto the field point, and is the layered medium Green’s function. Such function is

defined as the scattered x-directed field at the field point for a unit-strength x-directed

electric source at the source point, and it can be theoretically written as (Slob and

Fokkema 2002):

(2)

(3)

where subscript TM and TE stand, respectively, for transverse magnetic mode and transverse

electric mode. The R global reflection coefficients describe the reflected part of the wave in

each mode. is the vertical wave number, where subscript n represents the

number of layers and it is equal to 1 in equation (3) (index of the top layer (free space)),

and . The global TM-mode and TE-mode reflection coefficients at

the n = 1…N interface are given by:

(4)

(5)

(6)

(7)

The antenna characteristic functions and can be determined through a

proper calibration of the antenna, involving the measurement of the raw radar signal for

particular antenna-medium configurations, for which the value of the Green’s function can be

calculated (e.g., measurements with the antenna at different heights over a copper sheet). In

9
particular the coefficient, i.e., the global reflection of the antenna in free space, can be

derived by performing a radar measurement with the antenna pointed toward the sky. It is worth

noting that once the antenna factors are determined, they can be considered as constant

regardless from the type of medium investigated. Accordingly, all the antenna effects can be

filtered out from the raw radar data in order to retrieve the value of the Green’s function

Model inversion

This method relies on a comparison between the value of the Green’s function measured on the

test samples and the value of a theoretically modeled Green’s function ,

retrieved through the aforementioned theoretical assumptions.

It is therefore possible to model the behavior of the Green’s function depending on fixed

electromagnetic and geometrical parameters, such as the relative dielectric permittivity , the

electrical conductivity , the wave number , the number of layers N illuminated by the EM

wave, the thickness of the nth layer, and the distance from the antenna phase centre to the

surface of the investigated medium. A mono-layered configuration as well as a fixed distance of

0.32 m between the soil sample surface and the antenna aperture are considered in this study,

with the equivalent point source in turn being located at 0.07 m from the antenna aperture.

It is then possible to build a field of dielectric parameter values, sized [P×M], with P and M

being, respectively, the dimensions of the dielectric permittivity and the electrical conductivity

vectors taken into account. Accordingly, a 3D matrix of Green’s function values

can be produced in the evaluated frequency range, with dimensions [P×M×F], where

is the number of collected frequencies, being B and , respectively, the selected frequency

bandwidth and the frequency step.

10
The inversion process consists in retrieving the dielectric parameters and of the analyzed

materials starting from the measured radar data, expressed in terms of rather

than , and from the theoretically expected behavior of the modeled Green’s function

. These two unknown values are returned through a search for the best matching

between measured and modeled Green’s functions. In this regard, by defining a vector

, the wanted value of b is the one minimizing an objective function

expressing the error between measured and modeled Green’s functions. In this study, the

objective function is computed as follows (Patriarca et al. 2011):

(8)

Frequency dependence of conductivity

In the GPR frequency range, a frequency-dependent behavior of the material apparent

conductivity (including dielectric losses) usually occurs. This is mainly due to relaxation

mechanisms and Maxwell-Wagner effects in the soils. In this regard, it is worth mentioning the

locally linear relationship between and f proposed by Lambot et al. (2004a):

(9)

where is the reference value of conductivity at the starting frequency fi, which can be

evaluated through the inversion process, and is the variation rate of the . Therefore, when

considering , namely, the relevant conductivity for the nth layer, variable in the range [10-1 ÷

11
10-4 Sm-1], the parameters affecting the objective function must be taken into account in a new

vector . Typical values of lie in the range [10-10÷10-12 Ssm-1]. In this work, a

constant value of a equals to 1×10-10 Ssm-1 is adopted, so that the sizes of vector b are related

only with two dielectrics. It is worth mentioning how Lambot et al. (2004a) have found the

linear relationship of Equation (9) to be very suited for the 1 – 2 GHz frequency range, while

higher discrepancies of this model were observed for wider frequency intervals.

Implementing the Full-Wave Inversion method with the Time Domain Signal Picking

approach

This approach has been adopted according to the results from Tosti et al. (2013), wherein

laboratory tests were carried out on soil samples with same dimensions than those used in this

work. In particular, a good effectiveness was proved in forecasting dielectric permittivity values

by estimating the two-way travel time of the measured radar signal within the medium, being

such estimates comparable to those coming from the application of Full-Wave Inversion (FWI)

technique.

The method provides the application of an Inverse Fast Fourier Transform (IFFT) to convert the

domain of the signal measured from spectral into time. Once having the transformed signal

, the time delay between the air-medium and the medium-perfect electric

conductor (PEC) reflections can be measured, by knowledge of the thickness h1 of the

formworks, and the wave velocity through the medium v1 estimated by . It is

therefore possible to retrieve the value of permittivity by letting , where c is

the speed of light in free space. From now on, such approach will be referred to as the Time

Domain Signal Picking (TDSP) technique. Figure 1 shows a typical scenario of measured signal

where the two aforementioned peaks of reflection are clearly identified.

12
FIGURE 1. Measured Green’s function in the time domain and time delay

between air-medium and medium-PEC interfaces.

The implementation of the TDSP technique within a FWI approach subsequently provides to

retrieve a first approximation value of dielectric permittivity for each sample. As

represented in the flowchart of Figure 2, this value can therefore be used as a starting point for a

deeper analysis in the spectral domain, to be further refined for retrieving information about the

other electromagnetic properties. The conventional FWI approach is then employed over a

significant range of dielectric permittivity values, in the neighborhood of the value retrieved by

the TDSP approach, namely, within the range ± 10% of the value.

FIGURE 2. Flowchart representing the implementation of the TDSP technique within a FWI

approach.

Surface Reflection Method

The use of the Surface Reflection Method (SRM) leads to the evaluation of the relative

dielectric permittivity of a mono-layered system by comparing the amplitude of the air-soil

interface reflection and the amplitude from the air- PEC interface, taken as a reference. Such

approach is commonly employed when measuring with high-frequency off-ground GPR

systems. (Davis and Annan 2002, Serbin and Or 2004).

When considering one soil characterized by a certain value of permittivity , the reflection

coefficient R at the surface separating air and soil layers can be determined as follows (Redman

et al. 2002):

(10)

13
where 1 is the value of the dielectric permittivity of the air. Equation (10) relies on the twofold

assumptions of i) a negligible conductivity and ii) a simplified scenario of air layer over a

homogeneous half space (Lambot et al. 2006). By considering that the magnitude of the

reflection coefficient R at a given height position h above the surface can be also expressed as

the ratio between the amplitude of the reflection from the ground surface to the amplitude

of the reflection over a copper shield or a metal plate (with a reflection coefficient of -1),

both measured at the same height h from the ground, it is possible to calculate the permittivity

of the soil as:

(11)

This approach has the main advantage of providing an estimation of the soil permittivity

without knowing the subsurface reflector position or the wave propagation velocity through the

medium, as in case of TDSP applications. Nevertheless, it is clear how an inhomogeneous water

distribution within the medium could lead to different results depending on the method applied.

Indeed, for higher particle-sized materials, a gravimetric behavior is prevalent for the water,

which tends to stratify at the bottom as a water table. This water content clearly influences

results of FWI or TDSP methods, while it could be neglected by SRM if the water table is too

deep with respect to the air-soil interface. Many literature studies have dealt with the capability

of the SRM to sense water content on the basis of soil physical properties. Huisman et al. (2003)

considered such influence as relevant up to depths of 20 cm. Conversely, Serbin and Or (2004)

narrowed the reliability of this method to the first 1 cm below the soil surface. Therefore, it

seems clear how this uncertainty can be considered as the main drawback of the SRM (Lambot

et al. 2004b). Further drawbacks include a high dependence of this technique on surface

roughness, along with a high sensitivity to moisture for lower water contents (Davis et al. 1994).

On the other hand, it is also worth citing the remarkable advantage of a significant lightness in

computational requirements.

14
Volumetric Mixing Formulae

Volumetric Mixing Formulae (VMF) rely on the volumetric fraction and the dielectric

permittivity of each component of a multi-phase medium, and enable to assess the volumetric

water content (Dobson et al. 1985, Roth et al. 1990).

By considering a medium with n-phases, the general expression of a VMF can be written as

follows:

(12)

where represents a geometrical fitting parameter depending on the inner structure of the

medium (Lichtenecker and Rother 1931), while and stand, respectively, for the volume

fraction and the dielectric permittivity of the component. By implementing Equation (12) on

a three-phase system and by knowledge of the porosity of the soil material, it is therefore

possible to determine the permittivity of the medium:

(13)

where , and are, respectively, the permittivity of free water, of the solid matrix and of

the gaseous phase, and θ stands for the volumetric water content of the multi-phase system.

According to this, the permittivity of the solid matrix can be evaluated by considering the

relative permittivity of the multi-phase medium in dry conditions ( , as follows

(Patriarca et al. 2013):

(14)

In case of clayey soils, it is necessary to consider a four-phase medium and, accordingly, a VMF

expressed as follows (Roth et al.1990):

(15)

where and stand for the volumetric content and the permittivity of the clay fraction,

respectively. In addition, the permittivity of the soil matrix can be determined by considering

the dry conditions of the system (Patriarca et al. 2013):

15
 (16)

In this paper, values of were retrieved from the work of Tosti et al. 2013, wherein

laboratory tests on the same materials were carried out in dry conditions. Many efforts were

devoted in the past to characterize the value of , which can vary between -1 and 1. By letting

, it is assumed to have a wave travel time through the medium equal to the travel times

within each singular component weighted by the volume (Birchak et al. 1974, Dobson et al.

1985, Gorriti and Slob 2005). Roth et al. (1990) proposed a value of equals to 0.46 for a

three-phase medium, while Dobson et al. (1985) retrieved a value of 0.65 for a four-phase

mixture. An interesting work has been recently developed by Patriarca et al. (2013), who

developed a method for determining the optimal value for different clay-rich mixtures. In this

work, a value of 0.5 is adopted for the factor.

Among the main drawbacks of this method, it can be mentioned the need to have an a-priori

knowledge of some physical properties of the multi-phase medium, as well as of the dielectric

permittivity of each component.

Rayleigh Scattering Method

A recent approach relying on signal processing in the frequency domain was proposed by

Benedetto (2010). The main advantage of the Rayleigh Scattering Method (RSM) consists in

avoiding core sampling to calibrate the system, since the volume fractions of the three phases in

the medium are not accounted for. In this regard, can be directly estimated by frequency

analyses of the GPR signal without estimating the dielectric permittivity. The main assumption

are that in the unsaturated domain, electromagnetic waves are scattered by water droplets

(Drude 1902), thereby a shifting of the central frequency of the waves occurs (Bohren and

Huffman 1983).

16
Rayleigh scattering is traditionally used to explain the shifting of the frequency of the scattered

signals. A shift in the frequency distribution of the reflected signals has already been observed

in the past; however, the cause of such shift was not identified or deeply investigated (Narayana

and Ophir 1983; Ho, Gader, and Wilson 2004).

Overall, it is well-known that scattering is generated by singularities or non-homogeneities in

electromagnetic impedance. The process can be described as Rayleigh scattering whether the

dimensions of these non-uniformities are much smaller than the wavelength of the EM wave.

Analytically, the size of a scattering particle is defined by the ratio x = 2πr / λ, where r is the

radius of the particle, and λ represents the wavelength of the signal. According to this, Rayleigh

scattering occurs in the small size parameter regime when x ≪ 1. Scattering from larger

spherical particles was explained by Mie (1908) for an arbitrary size parameter x. When small

values of x are considered, the Mie theory falls in the Rayleigh approximation.

By means of both several assumptions on the three-phase porous medium properties and

simplifications of the physics, Benedetto (2010) defined the following formulation:

2
    

4
  r      1 
2 2  (17)
 
1  cos2  s  2f r     
 
6
  d 
2
I  s , f   I 0  f    1 f 
2 2 
2R  c0  1 f        2
   r       2
2 2 
  1 f   

whit R being the distance between the observer and the particle, θs represents the angle of

scattering, f is the frequency of the electromagnetic signal, c0 is the velocity of free space, μr is

the magnetic permeability, ε∞ is the dielectric constant of the full-polarized medium at an

infinite frequency electromagnetic field, Δε = εstatic−ε∞ is the difference between the permittivity

values of, respectively, a steady and an infinite frequency electromagnetic field, τ is the

relaxation time, and d stands for the diameter of the particle. According to the water content, a

non-linear modulation of the electromagnetic signal is produced by scattering. In this regard, the

author demonstrated how the peak of frequency was a comprehensive indicator, negatively

related to moisture. Accordingly, since scattering is caused by water presence in the medium,

more scattering events are expected as the water content increases.

17
In line with Equation (17), it was observed how the several frequency components of the

frequency spectra were differently scattered, depending on the soil type and water contents, with

the peak of the frequency spectrum fP being the frequency component having the maximum

scattered intensity of the EM wave. On the basis of several experimental evidences, it was

proposed the following regression law for determining water content θ, expressed in %, from

the value of the peak of frequency fP, expressed in Hz×108:

  A  f P  B (18)

where A and B are regression parameters calibrated by means of laboratory tests on different

soil samples. In this paper, the RSM was applied to analyse the behavior of clay-free and clay-

rich soil samples and to provide further insights about water content evaluation in the soil types

taken into account herein, according to Equation (18).

EXPERIMENTAL FRAMEWORK

Experimental design

The main purpose of the experimental design has been to provide a research scenario capable to

sufficiently represent the electromagnetic behavior of clayey soils, outlined by the combination

of different-textured soil samples in both clay-free and clay-rich conditions along with water

contents from dry to saturation. A detailed description of the samples preparation protocols

followed in this work, including compaction procedures and clay mixing, can be found in Tosti

et al. 2013. In addition, standard procedures for cross-checking the homogeneous wet

conditions within the soil samples have been implemented herein, as described later.

Test devices and equipment

Experimental tests were carried out using two different GPR systems. An ultra wideband

(UWB) stepped frequency continuous wave (SFCW) radar, was set-up by using a vector

network analyzer (VNA) manufactured by Hewlett Packard (8753C, 300 kHz – 3 GHz) to

18
collect data in the frequency domain (Figure 3a). More information on the antenna properties,

the calibration procedures, the experimental setup for such type of off-ground measurements, as

well as about the dimensions of the test formworks can be found in Tosti et al. 2013.

In addition, a pulsed radar system with ground-coupled antennas (pulseEKKO PRO

manufactured by Sensors & Software Inc., Canada), 500 MHz center frequency of investigation,

was used in a bi-static configuration and common offset (Figure 3b). Data were collected in the

time domain, using a 40ns time window and a time step of 5×10-2 ns.

FIGURE 3. Test devices and equipment a) SFCW radar set-up using a vector network analyzer

HP 8573C (Hewlett Packard Company, USA) and a linear polarized double-ridged broadband

TEM horn BBHA 9120 A (Schwarzbeck Mess-Elektronik, Germany), in a mono-static

configuration. b) PulseEKKO PRO 500 pulsed radar system, manufactured by Sensors &

Software Inc., Canada.

Materials and laboratory testing

Typical road materials used for unbound pavement layers construction and subgrade soils were

used for laboratory testing. In more details, three different soil types, classified by the AASHTO

soil classification system as A1, A2, and A3 were considered, being, respectively, gravel (A1,

grain size 4-8 mm), coarse sand (A2, grain size 1-2 mm), and fine sand (A3, grain size 0.125-

0.250 mm).

Concerning the clay-rich samples, an amount of 15% by weight of bentonite clay was added

within the above three types of undisturbed materials. The Colclay A90 bentonite clay,

manufactured by Ankerpoort NV, The Netherlands, is mineralogically mainly composed by

smectite, consisting of silica (62% by weight) and alumina (21% by weight) sheet-like bounded

particles. Mineral thin units get separated by water intrusion between the particles, thereby

causing a sudden dispersion of the bentonite in the water.

19
Increasing amounts of water were gradually added to the soil samples from dry up to saturated

conditions. After the mixing and compaction procedures, the bulk density of each soil sample

was collected. To ensure homogeneous moisture conditions within the samples, a number of

three samplings was carried out in the barycenter and in the lower and upper corners of each

sample, alongside the diagonal line of the specimens, such that it was possible to determine the

average volumetric water content of each sample after drying in the oven the sampled material.

A relatively high moisture homogeneity was broadly verified in all the samples according to the

low values of standard deviation determined between the aforementioned three sampled

amounts of VWC (i.e., standard deviation σθ < 2%). A thorough list of the main physical

properties of the dry-member materials used in this work, including bentonite clay, can be found

in Tosti et al. 2013.

RESULTS AND DISCUSSION

Soil behavior analysis through permittivity-based methods

According to the flowchart of Figure 2, permittivity values from the TDSP technique were

firstly estimated from the measured signal in time domain, after implementing the IFFT of the

responses measured in the frequency domain.

In a second step, the inversion process is run. According to this, each measured signal

is compared with a 3D matrix of Green’s functions , sized [P×M×F],

being P = 75 (i.e., ), M = 60 (i.e., ), and F = 640 (i.e., number of frequencies collected),

being n = 1, with a relatively large parameter space used for running such inversions. As

previously mentioned, the target value of can be searched in the parameter space defined by

± 10% of the value of first approximation. Furthermore, the frequency-dependent

electrical conductivity has been evaluated according to Equation (9), being the parameter

space for variable in the range [10-1÷10-4 Sm-1] within a bandwidth B spanning the 1 – 3 GHz

20
frequency range. Finally, the F parameter has been obtained by the ratio between the above

cited bandwidth B and a frequency step fs of 3.125 MHz.

Figure 4a,b depicts two examples of measured and modeled responses in the frequency and time

domain by FWI application.

FIGURE 4. Measured and modeled Green’s functions in the frequency (amplitude and

phase ) and time ( ) domain. (a) soil sample 23 (A3, clay = 0%, θ = 10.9%); (b) soil

sample 45 (A3, clay = 15%, θ = 9.7%).

As it can be seen in the plots, a smoother behavior can be observed in case of clay

presence (Figure 4b), being the same soil type (i.e., A3) considered in both figures, with

comparable volumetric water contents. Amplitudes for the clay-free A3 soil sample (Figure 4a)

reach higher peak values than in the presence of clay. Concerning the Green’s functions in the

time domain, two main peaks of reflection can be clearly singled out in both these figures.

It is also worth noticing how signal amplitudes in the time domain can be affected by

mismatches occurring in the frequency domain. The comparison between the two plots of

Figure 4a,b for the clay-free and the clay-rich soil samples shows the positions of the first main

peaks of reflection (namely, the air-soil interfaces at the subsequent zero-amplitude positions

along the x axis) at 2.03 ns and 2.02 ns, respectively. Coherently, the second main peaks of

reflection (namely, the soil/PEC interfaces at the subsequent zero-amplitude positions along the

x axis) are measured at 3.63 ns and 3.43 ns, thereby providing time distances = 1.60 ns for (a)

and = 1.40 ns for (b). Such trend can be considered a comprehensive case study for the

overall behavior of the road materials investigated, being the position of the first main peak of

reflection approximately the same for all the samples, while the second main position usually

locates in floating time distances, as expected by the different physical conditions manufactured

in the specimens. Relatively good matches between permittivity estimates with TDSP and FWI

21
techniques have been found, whereby errors ≤ 8%, 7%, and 6% can be detected for A1, A2, and

A3 soil types, respectively.

In addition to the above two techniques for permittivity estimation, the SRM was used to

retrieve further values of dielectric permittivity from the measured signals in the time

domain representing each soil sample. On the other hand, permittivity values from the

VMF approach were also inferred by exploiting the physical properties of the soil samples along

with the water contents gravimetrically determined.

Methods comparison

Table 1 lists the values of relative dielectric permittivity assessed by the above four methods in

both clay-free and clay-rich conditions. The overall trend proves how increasing values of

relative dielectric permittivity are reached for higher water amounts in samples. Moreover,

considerably lower values of permittivity can be observed in clayey samples, probably due to

the swelling properties of clay that tightly bound water particles by molecular forces preventing

a full polarization of the water dipoles when the EM field is applied. Much more than in the

case of clay-free samples, the four methods return similar permittivity values for the clay-rich

samples, whereby slighter variations of permittivity between dry-member and end-member (i.e.,

saturated) cases are also encountered. Considerable differences of dielectrics among all the four

approaches are instead detected in clay-free cases.

TABLE 1. Relative dielectric permittivity values retrieved using different processing methods

for 0% and 15% clay samples.

The plots in Figure 5a-f provide interesting insights on the EM response of different textured

soils with varying clay and water contents. Petrophysical relationships from Topp, Davis, and

Annan (1980) were used here for best comparing the performances of the processing techniques

employed with the empirical estimates of relative dielectric permittivity based on the water

22
content measurements. Both the Topp’s general expression and the Topp’s site-specific

relationship for the Rubicon sandy loam (SL) soil type, with grain size distribution comparable

to the soil types analyzed herein and very low clay contents, were taken into account for the

analyses on the clay-free soil samples, while only the Topp’s general expression was considered

for the clay-rich member cases.

Concerning clay-free samples behavior (i.e., Figures 5a-c), it can be argued how the agreement

of the processing methods in θ estimate is highly dependent on the soil texture. In particular, the

lower is the grain size of the material, the similar and closer are the EM responses retrieved. To

broadly quantify the errors between permittivity estimates by different approaches, the

Normalized Root Mean Square Deviation (NRMSD) index has been therefore evaluated for

each pair combination of methods, as follows:

(19)

where and are the estimated permittivity values for, respectively, the considered and

the reference method, n is the number of parameters, and are, respectively, the

maximum and minimum permittivity values within the population of the reference method

taken into account. Table 2 lists the NRMSD values for the whole set of processing techniques

that have been compared each other. Such statistics confirm a good reliability in permittivity

estimates of the A3 soil type, with negligible errors. On the contrary, these errors increase for

higher grain sizes up to 0.84 (i.e., NRMSD value by FWI and Topp (general) comparison for

A1 soil samples).

TABLE 2. Normalized Root Mean Square Deviation (NRMSD) values in clay-free soil samples

by comparing the different permittivity-based methods.

23
Experimental tests have highlighted that the greater the grain size of samples, the more

remarkable the heterogeneities observed in water distribution. In case of the A1 soil type

already with relatively low moisture contents, the experimental observations have shown how a

free-water layer had begun to form at the bottom of the test box. This could be due to a very low

threshold of loosely bound molecular and capillary water, being the amount of capillary water

controlled by the soil texture, soil structure, organic matter and gravity (Lyon and Buckman

1937). Accordingly, capillarity tends to have a greater contribution in finer-textured materials,

and much more capillary meniscus around single or multiple particles are formed. This can be

singled out according to the trend of permittivity values retrieved by the FWI with respect to

SRM, VMF, and Topp-based relationships. When considering A1 soil samples (Figure 5a) in

the gravel size domain, capillarity occurs up to very low volumetric water contents. Indeed,

similar estimates can be seen up to θ = 2.68%, beyond which the FWI begins to return very

much higher values of permittivity than the other techniques. Useful insights about the

hydraulic and EM behaviors of similar gravel aggregates for increasing water contents can be

found in Scullion and Saarenketo (1997).

One main explanation for this behavior lies on the own theoretical principles of these processing

techniques. By experimental evidence, the SRM approach returns relatively flat values of

permittivity within the A1 and A2 soil samples, respectively in Figures 5a and 5b, with an

increasing trend observed for the last two member-cases of A2 soil samples when the

permittivity approaches the analogous estimates by the other processing methods. In line with

this, a more regular and slight increasing behavior can be also seen in A3 soil samples (Figure

5c), wherein major effects of capillarity occur. Within this framework, when water content

increases in coarser-sized materials such as A1 soils, the upper unsaturated volume of the

formwork retains the same levels of loosely bound water, while the thickness of the water layer

progressively increases at the bottom. Since the SRM strongly relies on surface reflections, the

higher is the grain size of the soil (i.e., lower amounts of loosely bound and capillary water), the

more different will be the values of permittivity retrieved with respect to the other processing

24
techniques considered herein. Such behavior diverts when the depth to water layer approaches

the surface, namely, for moisture conditions close to saturation, and the SRM approach becomes

more sensitive to free water, thereby returning higher dielectrics. Where loosely bound water

contribution and capillary effects are higher (i.e., finer-textured soils), the SRM returns

permittivity values more consistent with the water added to the samples (i.e., A3 soil in Figure

5c).

A certain weak sensitivity to water is instead observed for permittivity estimates using the VMF

method, which slightly rise for finer-textured materials. Relying on the theoretical assumptions

of a homogeneous distribution of the multi-phase soil components within each weighted volume

and same wave propagation velocity through the medium, the relevant permittivity estimates do

not reflect the real distribution of water within the formwork. Such behavior is more

emphasized in coarser materials, such as the A1 soil sample, than in A2 and A3 soils, where

capillarity effects contribute at more homogeneously distributing moisture within the multi-

phase volume of the specimen.

The FWI approach takes instead into account the whole thickness of the samples, and returns a

modeled signal which includes all the possible information of the wave related to a non-

homogeneous distribution of water within the depth domain investigated.

On the other hand, the EM behavior in clay-rich soil samples (i.e., Figures 5d-f) is strongly

related to the presence and amount of clay particles, which deeply affect how the water

molecules adhere to the particle surface. In more details, the water dipole is oriented according

to the electrical charge of the clay particle when applying an EM field, by virtue of molecular

forces (Benedetto 2010). In this case, a high-density layer of tightly bound water around the

particle surface is formed. When water content increases, a ticker film of oriented dipoles

creates, and the outer water changes into the so-called loosely bound layer, since the bonding

forces decrease with the distance from the mineral surface. Basically, the aforementioned layers

of adsorbed water consist of monomolecular layers which surround negatively charged mineral

surfaces and further absorption water layers that can be tightly or loosely bound (Mitchell

25
1992). In this regard, when clay particles are present and water molecules are bound in the form

of adsorbed water, the EM field applied by GPR enables to orientate only the loosely bound

water dipoles. When increasing moisture contents, a ticker film of water around the soil

particles occur, thereby enabling the polarization of the outer dipoles. This reflects in a lower

variability of the soil dielectric properties with water content, regardless of the amount of θ.

A very good consistency in permittivity estimates is observed by the application of the FWI and

SRM methods within all the types of soil investigated, being the NRMSD indexes equal to 0.12,

0.16 and 0.21 for, respectively, A1, A2 and A3 soil samples (Table 3). This is due to fairly

homogeneous water content conditions throughout the whole thickness of the formwork, such

that lower differences between near surface permittivity values by SRM and full-depth

dielectrics by FWI occur. In addition to this, it is worthwhile to note how the application of the

VMF approach mostly overestimates the values (i.e., A2 and A3 soils in Figures 5e and

5f), since the water content contribution is completely taken into account by the theoretical

model, while the electrical losses due to the rates of tightly bound water to clay particles are not

considered.

Overall, the application of the Topp’s general relationship returns higher values of dielectric

permittivity for all the three types of soil. The VMF approach provides in turn intermediate

permittivity estimates between the Topp’s approach and the FWI and SRM techniques. Finally,

it should be noted how the best agreement among the various approaches is verified for the A1

soil type (Figure 5d).

FIGURE 5. Plots of volumetric water contents θ vs dielectric permittivity values by different

permittivity-based methods in clay-free – A1 (a), A2 (b), A3 (c) – and 15% of clay conditions –

A1 (d), A2 (e), A3(f) –.

26
TABLE 3. Normalized Root Mean Square Deviation (NRMSD) values in clay-rich soil samples

(15% of clay) by comparing the different permittivity-based methods.

Soil behavior analysis through the Rayleigh scattering method (RSM)

Radar traces from the pulseEKKO pulsed radar system were processed for each test. A

denoising step was applied by means of low-pass (i.e., 1000 MHz) and high-pass (i.e., 150

MHz) filters. Subsequently, a fast Fourier transform (FFT) was used for retrieving the frequency

spectrum of the radar signal, and the frequency peak fP of each spectrum, namely, the frequency

of the maximum amplitude, was then extracted. To enhance the accuracy of the processing, a

number of 20 traces per sample was collected, so that a more stable value of fP could be

achieved for each soil sample by averaging as follows:

(20)

with n being the total amount of i replicas, and fP,i being the frequency of the ith spectrum. Such

operation is useful in case of instability in the values of fP extracted by FFT within the same soil

sample, i.e., when the shift of the spectrum peak approaches to be sensitive to an increase of

water content. Further insights on the application of super-resolution techniques in the spectral

domain are discussed in Benedetto and Tosti (2013b).

An overview on the behavior describing the frequency modulations of signal spectra can be seen

in Figure 6a-f. Theoretical motivations and empirical observations on this phenomenon can be

found in literature for natural soils and compacted loose materials (e.g., Benedetto 2010), as

well as for concrete materials (e.g., Laurens 2005).

It is evident how the frequency peak moves systematically to lower values of frequency,

regardless of the grain size and the presence of clay, which however are fundamental to provide

a more comprehensive interpretation of such occurrence. Table 4 lists the measured values of

frequency peaks, case by case.

27
FIGURE 6. 3D representation of the frequency spectra modulation for the soil types

investigated from dry to saturated conditions. Clay-free conditions – A1 (a), A2 (b), A3 (c) –

and 15% of clay conditions – A1 (d), A2 (e), A3(f) –.

TABLE 4. Measured values of frequency spectra peaks ƒP [Hz × 108] for 0% and 15% clay

samples from dry to saturated conditions.

The overall behavior in both clay-free and clay-rich conditions in terms of peaks shifting is

represented, respectively, in Figure 7a and 7b. In addition and according to Equation (18), the

calibrated A and B parameters along with the relevant regression coefficients for volumetric

water content prediction are listed in Table 5. As it can be seen, the mean squares fitting linear

curves are characterized by relatively high correlation coefficients R2, especially for the finer-

grained soil types A2 and A3 with 15% of clay content.

FIGURE 7. Trend of values of frequency spectra peak (fP) across the range of moisture contents

investigated for clay-free (a), and clay-rich soil samples (b).

TABLE 5. Values of regression coefficients in Equation (18).

Numerical data show how the frequency shift occurs, across the range of moisture investigated,

with two main rates of displacement, whose dimensions vary according to the grain size and

clay content of the samples. Basically, a lower shift of frequency peaks is firstly observed (i.e.,

low amounts of water content). Secondly (i.e., increasing amounts of water content), more

remarkable displacements are noticed.

In both clay-free and clay-rich conditions, it can be noticed how the coarser is the grain size of

the soil the lower is the amount of moisture by which the aforementioned two steps of

displacement occur. Considering clay-free conditions (Figure 6a-c), the range of frequency

peaks variation ΔfP A1-0% clay for the A1 soil type reaches a water content of θ = 6.71% (Sample 7)

28
at the beginning of the second main step of displacement (relevant ΔfP A1-0% clay = 4.11×108 Hz).

The same step begins at θ = 21.43% (Sample 16) for the A2 soil type (relevant ΔfP A2-0% clay =

3.92×108 Hz), while lower displacements are reached for the A3 soil type, being the last

moisture value θ equals to 26.63% (Sample 27) with a significantly lower range of variation of

frequency peaks (relevant ΔfP A3-0% clay = 2.15×108 Hz).

In clay-rich conditions (Figure 6d-f), numerical results demonstrate how this occurrence is

broadly more attenuated, although its strong dependence on the grain size of the soil samples is

yet confirmed. Indeed, despite the lower ΔfP intervals, much higher volumetric water contents

than in clay-free conditions are observed within this range, and the finer is the grain size the

higher is the amount of moisture. In more details, such range of variation ΔfP A1-15% clay measures

3.14×108 Hz for the A1 clay-rich soil type, and it is reached in a wide field of moisture (θ =

23.95% - Sample 32). Within the A2 clay-rich soil samples, the highest ΔfP A2-15% clay measures

0.59×108 Hz with a relevant water content θ = 27.13% (Sample 41), while ΔfP A3-15% clay =

1.57×108 Hz and θ = 29.40% (Sample 51) are noticed for the A3 clay-rich finer soil type.

Such overall behavior can find reasonable explanations by relating the contribution of water,

clay and grain size of the undisturbed material. Several transition water states occur when

starting to add water in a system of dry soil particles, namely, tightly bound and loosely bound

adsorption water, capillary water, and free water (Mitchell 1992). According to this, different

rates of polarization of the water dipoles occur, with the grain size and mineralogy of soils

playing an important role in the modes of such occurrence (Saarenketo 1998). In case of A1

clay-rich soil samples (Figure 6d), it is found the widest experimental heterogeneity between

particles, i.e., among the undisturbed material (i.e., grain size 4-8 mm) and the bentonite clay.

When adding low amounts of water into a dry system composed by many particles from these

two types of population, both the gravel and the clay particles retain water by molecular forces,

therefore it is expected that the water dipoles are not fully-polarized by the EM field. In these

cases, lower variations of the frequency peaks of the spectra can be noticed. When increasing

the water content, clay particles continue to retain moisture, due to their swelling properties,

29
without undergoing any polarization, while gravel grains begin to loose water molecules, which

in contrast start to polarize. In such conditions, slightly higher variation of the frequency peaks

of the spectra can be observed. On the other hand, if the water amount added is considerable,

both the two types of population take part in the polarization of water dipoles, since a large

amount of clay particles begin to polarize all together. Such condition reflects in the highest rate

of variation of the frequency spectra peaks, as a result of a sudden amount of polarized water

dipoles. In case of finer-grained particles with 15% of clay content, such as the A3 clay-rich soil

samples (Figure 6f), the above two populations are instead characterized by particle sizes very

close each other, thereby the A3 fine sand is capable to retain higher amounts of water, as well

as losing more gradually water molecules. In line with this, such “shifting” behavior involving

the peaks of frequency spectra is much more attenuated.

When considering clay-free conditions (Figure 6a-c) with a unique population of soil particles

and a relatively homogeneous grain size, the shift of the frequency peaks of spectra occurs more

rapidly and linearly, although a higher disorder of the several frequency components of the

spectra can be broadly observed.

SUMMARY OF THE MAIN RESULTS

Permittivity-based methods

General considerations

 Higher water contents reflect into increasing values of relative dielectric permittivity.

 Considerable lower permittivity values in clayey soil samples are observed with respect to

clay-free samples.

 Similar permittivity values are returned with the four processing methods used in this study

in case of clayey samples. Slighter variations of permittivity between dry-member and end-

member cases are also encountered.

30
Clay-free samples

 The agreement of the processing techniques used in this study in θ estimate is highly

dependent on the soil texture: the lower is the grain size, the closer are the relative dielectric

permittivity values.

 The FWI technique has a higher sensitivity than the other permittivity-based processing

methods to single out the low threshold between loosely bound molecular and capillary

water in coarser-grained materials (i.e., A1 and A2 soil types) and determining reliable

values of moisture within the whole depth domain investigated.

 The SRM seems to be more suited for characterising the loosely bound water content of the

upper (surface) part of the samples in coarser-grained materials (i.e., A1 and A2 soil types)

when moisture conditions are not saturated.

 The VMF approach used herein does not provide worthwhile information for distinguishing

any transition water states due to the theoretical assumptions of the model itself, which

diverts from the real grain size structure of the soils investigated.

Clay-rich samples

 The Topp’s general relationship returns higher values of relative dielectric permittivity than

the processing techniques employed here for all the three types of soil.

 Good results consistency is observed between the FWI and the SRM methods within all the

three types of soil investigated due to the clay ability of retaining water and fairly

homogenising its content within the soil volume.

 In most cases, the VMF approach overestimates the permittivity values with respect to FWI

and SRM, since the rates of tightly bound water to clay content are not taken into account

by its theoretical model.

Rayleigh scattering method

31
 When increasing the water content, the peak of the frequency spectra of the signals moves

systematically to lower values of frequency, regardless of the grain size and the clay

content.

 In both clay-free and clay-rich samples, the highest linear correlation coefficients in

frequency peaks shifting vs VWC relationships are found for finer-grained clayey soil types

(i.e., A2 and A3)

 Two main rates of displacement for the shifting of the frequency spectra peaks are observed.

The occurrence and dimension of such behaviour vary according to the grain size and the

clay content of the soil samples.

 In both clay-free and clay-rich samples, the coarser is the grain size of the soil, the lower is

the amount of water whereby the frequency peaks displacement takes place.

 The frequency spectra peaks displacement is more attenuated in clay-rich than in clay-free

conditions. Anyhow, in clayey conditions, the wider is the grain size heterogeneity between

the undisturbed soil and the clay particles (i.e., A1 soil type), the higher is the rate of

frequency displacement.

CONCLUSIONS

This study is motivated by the need to understand the dielectric behavior of clayey subgrade

soils and unbound load-bearing layers using ground-penetrating radar. Data from two different

radar systems are analyzed at the laboratory scale of investigation for three types of soil

classified by AASHTO as A1, A2, and A3 in both clay-free and clay-rich conditions (15% by

weight of clay) under different water contents.

Several permittivity-based methods were used for data processing, namely, the time-domain

signal picking technique, the full-wave inversion method, the surface reflection method, and the

volumetric mixing formulae. An implementation of the time domain signal picking technique

within the full-wave inversion method has been also presented. Considerably lower values of

32
dielectric permittivity are observed in clayey samples with respect to clay-free conditions.

Permittivity estimates very close to each other are also detected between dry-member and end-

member (i.e., saturated) cases by the four approaches. On the contrary, considerable differences

of dielectrics are detected in clay-free cases. Comparisons between the above dielectrics and the

volumetric water contents of samples, gravimetrically determined, highlight considerable

differences among the full-wave technique and the other approaches. A very good capability of

this method in evaluating permittivity values of unsaturated coarse-grained materials in clay-

free conditions is shown. Overall, differences in permittivity estimates decrease for finer grain

sizes, such that the full-wave inversion traces relatively well the trend of the Topp’s general

relationship for the A3 soil type. The surface reflection method and the volumetric mixing

approach do not perform well in characterizing the overall volumetric water contents in 10.5 cm

thick formworks.

Concerning clay-rich soil samples, a very good agreement in the trend of the full-wave

inversion approach and the surface reflection method is noticed, due to the swelling properties

of clay, which creates a relatively homogeneous mixtures of the multi-phase components of the

soil samples, such that low changes of permittivity throughout the thickness of the formwork are

encountered. In such clayey conditions, the Topp’s general expression broadly overestimates the

values of soil permittivity. By a lower entity, considerable overestimates are also encountered

using the volumetric mixing approach.

The ability to detect clay in soils was also investigated using a Rayleigh-based scattering

technique, which relies on analyses carried out in the spectral domain of the GPR signal. The

results demonstrate a strong contribution of water, clay and grain size of the undisturbed

material on the occurrence of scattering phenomena. Higher variations in the range of frequency

spectra peaks are noticed in clay-free soil samples with respect to clay-rich conditions, being

this frequency-dependent behavior more rapid and linear, despite the overall higher disorder of

the several frequency components of the spectra. On the contrary, clay-rich soils exhibit simpler

spectral behaviors, and frequency shifts develop across higher amounts of water content, due the

33
swelling properties of clay. Soil-specific regression functions are proposed for estimating water

content as a function of the peaks of frequency spectra.

ACKNOWLEDGEMENTS

The research was supported by the ASTRI project, financed by Roma Tre University under the

Framework Program for “Progetto di internazionalizzazione della ricerca e della formazione

della ricerca”. A strong support was also provided by the Delft University of Technology, for

enabling the use of facilities, materials, and laboratory sites. The authors are thankful to Dr.

Karl-Heinz Wolf for permission to use the laboratory sites, as well as to Ing. Karel Heller and

Ing. Wim Verwaal, for their invaluable help and support.

This work also benefited from the network activities carried out within the EU funded COST

Action TU1208 “Civil Engineering Applications of Ground Penetrating Radar”.

34
References

Abdi, M.R., Sadrnejad, A.,. Arjomand, M.A., 2009. Strength enhancement of clay by

encapsulating geogrids in thin layers of sand. Geotext. Geomembranes. 27, 447-455.

Abusharar, S.W., Han, J., 2011. Two-dimensional deep-seated slope stability analysis of

embankments over stone column-improved soft clay. Eng. Geol. 120, 103-110.

Al-Qadi I.L. and Lahouar S. 2004. Use of GPR for thickness measurement and quality control

of flexible pavements. Journal of the Association of Asphalt Paving Technologists 73, 501–528.

American Association of State Highway and Transportation Officials (AASHTO), 2011.

Roadside Design Guide, 4th edition. AASHTO, Washington, DC.

Benedetto A. 2010. Water content evaluation in unsaturated soil using GPR signal analysis in

the frequency domain. Journal of Applied Geophysics 71, 26–35. doi:

10.1016/[Link].2010.03.001.

Benedetto A., Benedetto F. and Tosti F. 2012a. GPR applications for geotechnical stability of

transportation infrastructures. Nondestructive Testing And Evaluation 27(3), 253–262. doi:

10589759.2012.694884.

Benedetto A., Manacorda G., Simi A. and Tosti F. 2012b. Novel perspectives in bridges

inspection using GPR. Nondestructive Testing And Evaluation 27(3), 239–251.

Benedetto A. and Tosti F. 2013a. Inferring bearing ratio of un-bound materials from dielectric

properties using GPR: the case of Runaway Safety Areas. In: Proceedings of the Airfield and

Highway Pavement 2013 Conference, Los Angeles, California, USA, pp. 1336-1347. doi:

10.1061/9780784413005.113.

Benedetto A., D’Amico F. and Tosti F. 2014a. Improving safety of runway overrun through the

correct numerical evaluation of rutting in Cleared and Graded Areas. Safety Science 62, 326–

338. doi: 10.1016/[Link].2013.09.008.

Benedetto, A, Tosti, F., Pajewski, L., D’Amico, F., and Kusayanagi, W. 2014b. FDTD

Simulation of the GPR Signal for Effective Inspection of Pavement Damages. In: Proceedings

35
of the Fifteenth International Conference on Ground Penetrating Radar, Bruxelles, Belgium,

pp. 513-518.

Benedetto, A., Tosti, F., Ortuani, B., Giudici, M., and Mele, M. 2015. Mapping the spatial

variation of soil moisture at the large scale using GPR for pavement applications. Near Surface

Geophysics. 13 (3), 269–278. Doi: 10.3997/1873-0604.2015006.

Benedetto F. and Tosti F. 2013b. GPR spectral analysis for clay content evaluation by the

frequency shift method. Journal of Applied Geophysics 97, 89–96. doi:

10.1016/[Link].2013.03.012.

Beroya, M.A.A., Aydin, A., Katzenbach, R., 2009. Insight into the effects of clay mineralogy on

the cyclic behavior of silt–clay mixtures. Eng. Geol. 106, 154-162.

Birchack J.R., Gardner C. G., Hipp J.E. and Victor M. 1974. High dielectric constant

microwave probes for sensing soil moisture. Proceedings of the IEEE. Vol 62. No. 1.

Bohren CF, Huffman D. Absorption and scattering of light by small particles. New York: John

Wiley; 1983.

Daniels D.J. 2004. Ground Penetrating Radar. The Institution of Electrical Engineers, London.

Davis, J.L., Annan, A.P., 2002. Ground penetrating radar to measure soil water content, in

Dane, J.H. and Topp, G.C. (eds), Methods of Soil Analysis, Part 4, Soil Science Society of

America (SSSA), pp. 446-463.

De Benedetto, D., Castrignano, A., Sollitto, D., Modugno, F., Buttafuoco, G., Papa, G.L., 2012.

Integrating geophysical and geostatistical techniques to map the spatial variation of clay.

Geoderma, 171-172, 53-63.

Dobson, M.C., Ulaby, F.T., Hallikainen, M.T., El-Rayes, M.A., 1985. Microwave dielectric

behavior of wet soil. Part II. Dielectric mixing models. IEEE Transactions on Geoscience and

Remote Sensing 23, 35–46.

Drude, P., 1902. The theory of Optics. Longmans, Green, and Co, New York, 268–396.

Dudoignon, P., Causseque, S., Bernard, M., Hallaire, V., Pons, Y., 2007. Vertical porosity

profile of a clay-rich marsh soil. Catena 70, 480-492.

36
Fellner-Feldegg, H., 1969. Measurement of dielectrics in time domain. The Journal of Physical

Chemistry 73, 616–623.

Gómez-Ortiz, D., Martín-Crespo, T., Martín-Velázquez, S., Martínez-Pagán, P., Higueras, H.,

Manzano, M., 2010. Application of ground penetrating radar (GPR) to delineate clay layers in

wetlands. A case study in the Soto Grande and Soto Chico watercourses, Doñana (SW Spain).

Journal of Applied Geophysics, 72 (2), 107-113.

Gorriti A.G. and Slob E.C. 2005. Synthesis of all known analytical permittivity reconstruction

techniques of nonmagnetic materials from reflection and transmission measurements. IEEE

Geoscience and Remote Sensing Letters, Vol. 2, No. 4, doi: 10.1109/LGRS.2005.853199

Ho, K.C., Gader, P.D., Wilson, J.N., 2004. Improving landmine detection using frequency

domain features from ground penetrating radar. Proceedings. 2004 IEEE International

Geoscience and Remote Sensing Symposium, IGARSS '04, Vol. 3, pp. 1617–1620.

Huisman J.A., Hubbard S.S., Redman J.D. and Annan A.P. 2003. Measuring soil water content

with ground penetrating radar: a review. Vadose Zone Journal 2, 476–491. doi:

10.2113/2.4.476.

Huston D.R., Hu J., Maser K., Weedon W. and Adam C. 1999. Ground penetrating radar for

concrete bridge health monitoring applications. In: Proceedings of SPIE 3587, pp. 170-179.

doi:10.1117/12.339922.

Lambot S., Slob E.C., van den Bosch I., Stockbroeckx B. and Vanclooster M. 2004a. Modeling

of ground-penetrating radar for accurate characterization of subsurface electric properties. IEEE

Transactions on Geoscience and Remote Sensing 42, 2555–2568. doi:

10.1109/TGRS.2004.834800.

Lambot, S. , Rhebergen, J., van den Bosch, I., Slob, E.C., and Vanclooster, M. 2004b.

Measuring the soil water content profile of a sandy soil with an off-ground monostatic ground

penetrating radar, Vadose Zone Journal 3, 1063–1071. doi: 10.2113/3.4.1063.

37
Lambot S., Weihermüller L., Huisman J.A., Vereecken H., Vanclooster M. and Slob E.C. 2006.

Analysis of air-launched ground-penetrating radar techniques to measure the soil surface water

content. Water Resources Research 42(11), W11403. doi: 10.1029/2006WR005097.

Laurens, S., Balayssac, J-P., Rhazi, J., Klysz, G., Arliguie, G. 2005. Non-destructive evaluation

of concrete moisture by GPR: Experimental study and direct modeling. Materials and

Structures. 38(283), 827–832.

Lichtenecker, K., Rother, K., 1931. Die herleitung des logarithmischen mischungsgesetzes aus

allgemeinen prinzipien der stätionaren strömung. Physik Z. 32, 255–260.

Lyon, T.L., Buckman, H.O., 1937. The Nature and Properties of Soils. Macmillan, New York.

p. 391.

Mahmoudzadeh, M.R., André, F., van Wesemael, B., Lambot, S., 2011. Clay content and soil

moisture mapping using on-ground time-domain GPR. In Proceedings of the 2nd Workshop on

Proximal Soil Sensing, Montreal, Canada, May 15-18. p. 44-47.

Mie, G., 1908. Beiträge zur Optik trüber Medien, speziell kolloidaler Metallösungen. Annalen

der Physik 330, 377.

Minet J, Wahyudi A., Bogaert P., Vanclooster M. and Lambot S. 2011. Mapping shallow soil

moisture profiles at the field scale using full-wave inversion of ground penetrating radar data.

Geoderma 161, 225–237. doi: 10.1016/[Link].2010.12.023.

Miqueleiz, L., Ramirez, F., Seco, A., Nidzam, R.M., Kinuthia, J.M., Abu Tair, A., Garcia, R.,

2012. The use of stabilised Spanish clay soil for sustainable construction materials. Eng. Geol.

133-134, 9-15.

Mitchell, J.K., 1992. Fundamentals of Soil Behavior, 2nd edn. Wiley, New York. p. 437.

Narayana, P.A., Ophir, J., 1983. On the frequency dependence of attenuation in normal and fatty

liver. IEEE Transactions on sonics and ultrasonics SU30 6, 379–383.

Odeh, I.O.A., McBratney, A.B., 2000. Using AVHRR images for spatial prediction of clay

content in the lower Namoi Valley of eastern Australia. Geoderma 97, 237-254.

38
Pakbaz, M.S., Alipour, R., 2012. Influence of cement addition on the geotechnical properties of

an Iranian clay. Appl. Clay Sci. 67-68, 1-4.

Patriarca, C., Lambot, S., Mahmoudzadeh, M.R., Minet, J., Slob, E.C., 2011. Reconstruction of

sub-wavelength fractures and physical properties of masonry media using full-waveform

inversion of proximal penetrating radar. Journal of Applied Geophysics 74, 26–37.

doi:10.1016/[Link].2011.03.001

Patriarca C., Tosti F., Velds C., Benedetto A., Lambot S. and Slob E.C. 2013. Frequency

dependent electric properties of homogeneous multi-phase lossy media in the ground-

penetrating radar frequency range. Journal of Applied Geophysics 97, 81–88. doi:

10.1016/[Link].2013.05.003.

Redman J., Davis J., Galagedara L. and Parkin G. 2002. Field studies of GPR air launched

surface reflectivity measurements of soil water content. In: Proceedings of the Ninth

International Conference on Ground Penetrating Radar, Santa Barbara, California, USA. S.

Koppenjan and K. Lee, Eds. 4758, pp. 156-161.

Richard, G., Cousin, I., Sillon, J.F., Bruand, A., Guérif, J., 2001. Effect of compaction on the

porosity of a silty soil: influence on unsaturated hydraulic properties. Eur. J. Soil Sci. 52, 49-58.

Robinson, D.A., Phillips, C.P., 2001. Crust development in relation to vegetation and

agricultural practice on erosion susceptible, dispersive clay soils from central and southern Italy.

Soil Till. Res. 60, 1-9.

Robinson, D.A., Campbell, C.S., Hopmans, J.W., Hornbuckle, B.K., Jones, S.B., Knight, R.,

Ogden, F., Selker, J. Wendroth, O., 2008. Soil moisture measurement for ecological and

hydrological watershed-scale observatories: a review. Vadose Zone Journal 7 (1), 58–389.

Roth, K., Schulin, R., Fluhler, H., Attinger, W., 1990. Calibration of time domain reflectometry

for water content measurement using composite dielectric approach. Water Resources Research

26, 2267–2273.

Saarenketo, T., 1998. Electrical Properties of Water in Clay and Silty Soils. Journal of Applied

Geophysics 40(1–3), 73–88.

39
Saarenketo, T., Scullion, T., 2000. Road evaluation with ground penetrating radar. J. Appl.

Geophys., 43 (2), 119-138.

Scullion T., Lau C.L. and Chen Y. 1994. Pavement evaluations using ground penetrating radar.

International Conference on Ground Penetrating Radar. In: Proceedings of the Fifth

International Conference on Ground Penetrating Radar, Kitchener, Ontario, Canada, pp. 449-

463.

Scullion, T., Saarenketo, T., 1997. Using suction and dielectric measurements as performance

indicators for aggregate base materials. Transportation Research Record 1577, TRB, national

Research Council, Washington, D.C., pp. 37-44.

Serbin G. and Or D. 2004. Ground-penetrating radar measurement of soil water content

dynamics using a suspended horn antenna. IEEE Transactions on Geoscience and Remote

Sensing, Vol. 42, No. 8. doi: 10.1109/TGRS.2004.831693.

Slob, E.C., Fokkema, J.T., 2002. Interfacial dipoles and radiated energy, Subsurface Sens.

Technol. Appl., 3(4), 399 – 419.

Slob E.C., Sato M. and Olhoeft G., 2010. Surface and borehole ground-penetrating-radar

developments. Geophysics 75(5), A103-A120. doi: 10.1190/1.3480619.

Srasra, E., Bergaya, F., Fripiat, J.J., 1994. Infrared spectroscopy study of tetrahedral and

octahedral substitutions in an interstratified illite-smectite clay. Clays and Clay Minerals, 42 (3),

pp. 237-241.

Topp G.C., Davis J.L. and Annan A.P. 1980. Electromagnetic determination of soil water

content: measurements in coaxial transmission lines. Water Resources Research 16, 574–582.

doi: 10.1029/WR016i003p00574.

Tosti, F., Benedetto, A., Calvi, A. 2014a. Efficient air-launched ground-penetrating radar

inspections in a large-scale road network. In Proc. of the 3rd International Conference on

Transportation Infrastructure, Pisa, Italy, Apr. 2014, pp. 703–709.

Tosti, F., Adabi, S., Pajewski, L., Schettini, G., and Benedetto, A. 2014b. Large-scale analysis

of dielectric and mechanical properties of pavement using GPR and LFWD. In: Proceedings of

40
the Fifteenth International Conference on Ground Penetrating Radar, Bruxelles, Belgium, pp.

268-273.

Tosti F., Patriarca C., Slob E.C., Benedetto A. and Lambot S. 2013. Clay content evaluation in

soils through GPR signal processing. Journal of Applied Geophysics 97, 69–80. doi:

10.1016/[Link].2013.04.006.

Triantafilis, J., Lesch, S.M., 2005. Mapping clay content variation using electromagnetic

induction techniques. Comput. Electron. Agr. 46, 203-237.

Uzan, J., 1998. Characterization of clayey subgrade materials for mechanistic design of flexible

pavements. Transportation Research Record, (1629), pp. 189-196.

van der Kruk J. and Slob E.C. 2004. Reduction of reflections from above surface objects in

GPR data. Journal of Applied Geophysics 55, 271–278. doi: 10.1016/[Link].2004.02.002.

Viscarra Rossel, R.A., Cattle, S.R., Ortega, A., Fouad, Y., 2009. In situ measurements of soil

colour, mineral composition and clay content by vis–NIR spectroscopy. Geoderma 150, 253-

266.

Wagner, W., Blöschl, G., Pampaloni, P., Calvet, J.C., Bizzarri, B., Wigneron, J.P., Kerr, Y.,

2007. Operational readiness of microwave remote sensing of soil moisture for hydrologic

applications. Nordic Hydrology 38 (1), 1–20.

Wobshall, D. 1978. A frequency shift dielectric soil moisture sensor. IEEE Geosci. Electronics,

Vol. GE16, 112-118.

Wuddivira, M.N., Robinson, D.A., Lebron, I., Bréchet, L., Atwell, M., De Caires, S., Oatham,

M., Jones, S.B., Abdu, H., Verma, A.K., Tuller, M., 2012. Estimation of soil clay content from

hygroscopic water content measurements. Soil Science Society of America Journal, 76 (5), pp.

1529-1535.

41
List of captions

FIGURE 1 Measured Green’s function in the time domain and time delay

between air-medium and medium-PEC interfaces.

FIGURE 2. Flowchart representing the implementation of the TDSP technique within the FWI

approach.

FIGURE 3. Test devices and equipment a) SFCW radar set-up using a vector network analyzer

HP 8573C (Hewlett Packard Company, USA) and a linear polarized double-ridged broadband

TEM horn BBHA 9120 A (Schwarzbeck Mess-Elektronik, Germany), in a mono-static

configuration. b) PulseEKKO PRO 500 pulsed radar system, manufactured by Sensors &

Software Inc., Canada.

TABLE 1. Relative dielectric permittivity values retrieved using different processing methods

for 0% and 15% clay samples.

FIGURE 4. Measured and modeled Green’s functions in the frequency (amplitude and

phase ) and time ( ) domain. (a) soil sample 23 (A3, clay = 0%, θ = 10.9%); (b) soil

sample 45 (A3, clay = 15%, θ = 9.7%).

TABLE 2. Normalized Root Mean Square Deviation (NRMSD) values in clay-free soil samples

by comparing the different permittivity-based methods.

42
FIGURE 5. Plots of volumetric water contents θ vs dielectric permittivity values by different

permittivity-based methods in clay-free – A1 (a), A2 (b), A3 (c) – and 15% of clay conditions –

A1 (d), A2 (e), A3(f) –.

TABLE 3. Normalized Root Mean Square Deviation (NRMSD) values in clay-rich soil samples

(15% of clay) by comparing the different permittivity-based methods.

FIGURE 6. 3D representation of the frequency spectra modulation for the soil types

investigated from dry to saturated conditions. Clay-free conditions – A1 (a), A2 (b), A3 (c) –

and 15% of clay conditions – A1 (d), A2 (e), A3(f) –.

TABLE 4. Measured values of frequency spectra peaks ƒP [Hz × 108] for 0% and 15% clay

samples from dry to saturated conditions.

FIGURE 7. Trend of values of frequency spectra peak (fP) across the range of moisture contents

investigated for clay-free (a), and clay-rich soil samples (b).

TABLE 5. Values of regression coefficients in Equation (18).

43
Figures

FIGURE 1 Measured Green’s function in the time domain and time delay

between air-medium and medium-PEC interfaces.

44
FIGURE 2. Flowchart representing the implementation of the TDSP technique within the FWI

approach.

45
FIGURE 3. Test devices and equipment a) SFCW radar set-up using a vector network analyzer

HP 8573C (Hewlett Packard Company, USA) and a linear polarized double-ridged broadband

TEM horn BBHA 9120 A (Schwarzbeck Mess-Elektronik, Germany), in a mono-static

configuration.

46
FIGURE 3. Test devices and equipment b) PulseEKKO PRO 500 pulsed radar system,

manufactured by Sensors & Software Inc., Canada.

47
FIGURE 4. Measured and modeled Green’s functions in the frequency (amplitude and

phase ) and time ( ) domain. (a) soil sample 23 (A3, clay = 0%, θ = 10.9%)

48
FIGURE 4. Measured and modeled Green’s functions in the frequency (amplitude and

phase ) and time ( ) domain. (b) soil sample 45 (A3, clay = 15%, θ = 9.7%).

49
FIGURE 5. Plots of volumetric water contents θ vs dielectric permittivity values by different

permittivity-based methods in clay-free – A1 (a), A2 (b), A3 (c) – and 15% of clay conditions –

A1 (d), A2 (e), A3(f) –.

50
51
FIGURE 6. 3D representation of the frequency spectra modulation for the soil types investigated from dry to

saturated conditions. Clay-free conditions – A1 (a), A2 (b), A3 (c) – and 15% of clay conditions – A1 (d), A2 (e),

A3(f) –.

52
53
FIGURE 7. Trend of values of frequency spectra peak (fP) across the range of moisture contents

investigated for clay-free (a), and clay-rich soil samples (b).

54
 Tables

TABLE 1. Relative dielectric permittivity values retrieved using different processing methods

for 0% and 15% clay samples.

Soil Soil
sample ID θ [%] εTDSP εFWI εSRM εVMF sample ID θ [%] εTDSP εFWI εSRM εVMF
(0% clay) (15% clay)
A1 A1
Sample 1 0.00 2.53 2.65 3.46 3.12 Sample 28 0.00 3.97 4.02 3.89 4.01
Sample 2 2.68 3.64 3.50 3.02 3.55 Sample 29 8.42 4.32 3.98 3.56 5.36
Sample 3 4.44 5.34 5.30 3.86 3.83 Sample 30 18.27 5.67 5.35 5.41 6.93
Sample 4 4.26 6.42 6.49 4.28 3.80 Sample 31 21.69 6.00 5.68 6.88 7.48
Sample 5 5.11 8.46 8.50 3.53 3.94 Sample 32 23.95 8.63 8.78 8.89 7.84
Sample 6 5.73 10.43 10.52 3.76 4.03 Sample 33 27.11 9.04 8.82 11.52 8.35
Sample 7 6.71 12.9 13.22 4.62 4.19 Sample 34 29.44 12.21 12.28 11.13 8.72
Sample 8 8.34 15.96 16.20 2.63 4.45 Sample 35 31.76 12.30 12.37 13.11 9.09
Sample 9 12.18 20.09 20.59 27.74 5.07

A2 A2
Sample 10 0.00 2.74 2.78 2.21 3.33 Sample 36 0.00 4.07 4.06 4.05 4.03
Sample 11 3.62 3.96 3.73 2.94 3.91 Sample 37 9.71 3.91 3.75 3.86 5.58
Sample 12 6.06 4.60 4.52 3.00 4.30 Sample 38 17.99 4.36 4.41 3.02 6.91
Sample 13 7.27 5.93 5.55 3.70 4.50 Sample 39 20.96 4.72 4.80 4.07 7.38
Sample 14 10.48 8.79 8.44 3.56 5.01 Sample 40 24.89 5.47 5.58 4.79 8.01
Sample 15 14.71 11.07 10.69 3.36 5.69 Sample 41 27.13 6.00 5.91 6.28 8.37
Sample 16 21.43 14.44 14.02 3.41 6.76 Sample 42 30.29 6.75 7.16 8.39 8.88
Sample 17 24.68 17.79 17.91 4.47 7.28 Sample 43 33.20 6.85 7.35 9.05 9.34
Sample 18 29.44 21.10 20.60 11.46 8.05

A3 A3
Sample 19 0.00 2.64 2.66 2.44 3.26 Sample 44 0.00 3.43 3.38 2.76 3.68
Sample 20 2.80 3.22 3.04 2.28 3.71 Sample 45 9.69 4.02 4.03 4.21 5.23
Sample 21 5.55 3.96 3.80 3.24 4.15 Sample 46 15.46 3.38 3.37 3.01 6.15
Sample 22 8.07 4.84 4.66 3.13 4.55 Sample 47 18.10 3.69 3.65 3.08 6.57
Sample 23 10.95 5.87 5.52 4.45 5.01 Sample 48 21.08 4.02 3.88 3.92 7.05
Sample 24 13.61 6.14 5.69 3.97 5.44 Sample 49 23.79 4.18 4.07 4.17 7.48
Sample 25 17.42 8.06 7.83 4.61 6.05 Sample 50 26.00 4.42 4.69 4.03 7.84
Sample 26 20.87 11.54 11.27 5.67 6.60 Sample 51 29.40 4.65 4.84 4.44 8.38
Sample 27 26.63 15.19 14.91 7.07 7.52

55
TABLE 2. Normalized Root Mean Square Deviation (NRMSD) values in clay-free soil samples

by comparing the different permittivity-based methods.

Methods NRMSD
A1 A2 A3
FWI SRM 0.44 0.39 0.28
FWI VMF 0.44 0.56 0.24
SRM VMF 0.24 0.11 0.21
Topp Gen. FWI 0.84 0.24 0.03
Topp Gen. SRM 0.28 0.39 0.40
Topp Gen. VMF 0.28 0.36 0.38
Topp SL FWI 0.83 0.11 0.03
Topp SL SRM 0.33 0.39 0.41
Topp SL VMF 0.31 0.37 0.38

56
TABLE 3. Normalized Root Mean Square Deviation (NRMSD) values in clay-rich soil samples

(15% of clay) by comparing the different permittivity-based methods.

Methods NMRSD

A1 A2 A3
FWI SRM 0.12 0.16 0.21
FWI VMF 0.24 0.37 0.55
SRM VMF 0.24 0.42 0.54
Topp Gen. FWI 0.21 0.36 0.46
Topp Gen. SRM 0.19 0.35 0.47
Topp Gen. VMF 0.30 0.39 0.43

57
TABLE 4. Measured values of frequency spectra peaks ƒP [Hz × 108] for 0% and 15% clay

samples from dry to saturated conditions.

Soil Soil
sample ID θ [%] fP [Hz×108] sample ID θ [%] fP [Hz×108]
(0% clay) (15% clay)
A1 A1
Sample 1 0.00 6.65 Sample 28 0.00 5.68
Sample 2 2.68 5.87 Sample 29 8.42 5.51
Sample 3 4.44 5.48 Sample 30 18.27 5.29
Sample 4 4.26 5.28 Sample 31 21.69 5.09
Sample 5 5.11 4.50 Sample 32 23.95 2.54
Sample 6 5.73 4.70 Sample 33 27.11 2.54
Sample 7 6.71 2.54 Sample 34 29.44 2.35
Sample 8 8.34 2.54 Sample 35 31.76 2.35
Sample 9 12.18 2.15

A2 A2
Sample 10 0.00 6.46 Sample 36 0.00 5.87
Sample 11 3.62 5.87 Sample 37 9.71 5.87
Sample 12 6.06 5.28 Sample 38 17.99 5.68
Sample 13 7.27 5.09 Sample 39 20.96 5.48
Sample 14 10.48 5.09 Sample 40 24.89 5.28
Sample 15 14.71 4.70 Sample 41 27.13 5.28
Sample 16 21.43 2.54 Sample 42 30.29 5.87
Sample 17 24.68 2.34 Sample 43 33.20 5.68
Sample 18 29.44 2.15

A3 A3
Sample 19 0.00 6.65 Sample 44 0.00 6.07
Sample 20 2.80 6.26 Sample 45 9.69 5.68
Sample 21 5.55 5.87 Sample 46 15.46 5.87
Sample 22 8.07 5.28 Sample 47 18.10 5.87
Sample 23 10.95 5.09 Sample 48 21.08 5.87
Sample 24 13.61 5.87 Sample 49 23.79 5.69
Sample 25 17.42 5.28 Sample 50 26.00 5.67
Sample 26 20.87 5.28 Sample 51 29.40 4.50
Sample 27 26.63 4.50

58
TABLE 5. Values of regression coefficients in Equation (18).

Clay Content [%] Parameter Soil

A1 A2 A3
8
0 A × 10 6.79 6.43 6.64
7
B × 10 4.34 1.56 1.02
R2 0.84 0.95 0.77

15 A × 108 6.34 5.83 5.96

B × 107 1.23 0.10 0.07
2
R 0.71 0.98 0.99

59