Construction

and Building

MATERIALS

Construction and Building Materials 19 (2005) 763772

www.elsevier.com/locate/conbuildmat

Measuring layer thicknesses with GPR Theory to practice

I.L. AL-Qadi
a

a,*,1

, S. Lahouar

b,2

Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, 205 N Mathews MC 250, Urbana, IL 61801, USA
b
Institut Superieur des Sciences Appliquees et de Technologie de Sousse, Cite Taala, Ibn Khaldoun, Sousse 4003, Tunisia
Available online 2 August 2005

Abstract
Ground penetrating radar (GPR) technology has been used to assess pavement performance and structure for the past 30 years in
a variety of ways. Yet after all this time, the main issue remains: How well does GPR work and under what conditions? Results show
that GPR works well for some situations but not as well for others. It is not currently used on a routine basis by the Departments of
Transportation in the US mainly because of diculties encountered while interpreting GPR data. These diculties are generally
attributed to the fact that the GPR reected signals that are collected depend largely on the a priori unknown dielectric properties
of the structural materials. Additional diculties arise from the fact that physically GPR cannot detect layers unless they have sufciently dissimilar dielectric constants. In practice, GPR has been used primarily for pavement layer thickness estimation and moisture accumulation localization within the pavement layers. To improve GPR prediction capabilities, dierent data processing
techniques have been developed that use the GPR reected signal to estimate the dielectric properties of surveyed structures, thus
determining their thicknesses. Other signal processing techniques have also been used successfully to enhance the quality of the GPR
signal in order to increase the accuracy of the data interpretation results.

2005 Elsevier Ltd. All rights reserved.
Keywords: Non-destructive testing; Ground penetrating radar; Pavement layer thickness estimation

1. Introduction
Determining exible and rigid pavement layer thicknesses is important for pavement evaluation and provides important data for pavement management
systems (PMS). For example, accurate predictions of
pavement layer thicknesses are needed for overlay design, quality control/quality assurance, and structural
capacity estimation of existing pavements to predict
their remaining service life. Currently, most Department
of Transportation (DTO) agencies evaluate layer thickness and the properties of dierent pavement layers
through the destructive process of extracting pavement
cores. While this procedure provides relatively accurate
*

Corresponding author. Tel.: 217 265 0427; fax: 217 333 1924.
E-mail address: alqadi@uiuc.edu (I.L. AL-Qadi).
1
Founder professor of Engineering and Director of Advanced
Transportation Research and Engineering Laboratory.
2
Assistant professor.
0950-0618/$ - see front matter
2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conbuildmat.2005.06.005

thickness measurements, it is time consuming, hazardous, requires trac control, provides limited information (as cores are usually taken every 300 m), and
cannot be performed annually since it adds to the pavement distress by causing man-made defects.
Another approach to estimate pavement layer thickness is to use the deections measured by a falling
weight deectometer (FWD). Yet this technique is slow,
costly, and does not accurately predict thicknesses because its main purpose is, usually, to backcalculate the
moduli of the pavement layers knowing their thicknesses. Another non-destructive alternative for pavement thickness estimation is to use GPR, which is
rapid, cost eective, and allows pavement surveys to
be conducted more eciently without disturbing either
the pavement structure or the highway trac.
During the past three decades, GPR has been used in
many studies for the non-destructive evaluation of
pavements. In rigid pavements, GPR has proven to be

764

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

feasible in locating dowels [1] and in detecting voids or

loss of support under slabs [2]. In exible pavements,
electromagnetic (EM) waves are found to serve as a tool
to detect moisture in the hot-mix asphalt (HMA) layer
and to locate moisture in the base layer that may lead
to structural damage [3,4]. Currently, the majority of
GPR applications in pavements are focused on determining layer thicknesses. For this application, dierent
researchers have reported varying performance levels
for the GPR tool, depending on the surveyed pavement
structure and the data analysis technique used. For example, Maser [5] reported a thickness accuracy of 7.5% for
hot-mix asphalt (HMA) layers ranging from 51 to
500-mm thick, and 12% for granular base layers ranging from 150 to 330-mm thick. This GPR performance
evaluation was based on comparisons between the thicknesses predicted from the GPR data and the thicknesses
measured from cores. In another study, Lahouar et al.
[6] used GPR to assess the condition of a four-lane,
17-mile section of Interstate I-81 in Virginia, in both
the northbound and southbound directions. The authors
reported a 6.7% error for predicting the HMA thickness
of a pavement approaching its service life. In contrast,
GPR measurements of the HMA layer of a one-year
old pavement (concrete and exible pavement sections)
at the Virginia Smart Road [7] were reported to have
an error of only 3.5% [8]. Al-Qadi et al. [9] used GPR
as a quality control/quality assurance (QC/QA) tool to
check the layer thicknesses of a newly built pavement
(Route 288, located near Richmond, VA). The study
reported a mean thickness error of 2.9% for HMA layers
ranging between 100 and 250 mm in thickness.
This paper summarizes the theoretical background
that could be used to estimate pavement layer thicknesses from GPR data. This theoretical analysis is then
validated by the presentation of eld data that evaluates
the accuracy of the GPR results and nds the conditions
that maximize its performance. In addition, alternative
analysis techniques that could be applied to increase
the GPR accuracy when the optimum conditions are
not met are also discussed.

2. Layer thickness estimation from GPR data

The principle of the GPR system used in this study
(impulse radar) is based on sending an EM pulse
through an antenna to the pavement surface and then
recording the reected pulses from the internal interfaces, where there is a contrast in the dielectric properties, as depicted in Fig. 1. The time dierence
measured between the reected pulses (i.e., t1 or t2)
can be used in conjunction with the dielectric properties
of the surveyed layer to determine its thickness. The
thickness of the ith layer could be computed according
to the following equation [10]:

Fig. 1. Typical GPR reections from a pavement system.

cti
d i p ;
2 er;i

where di is the thickness of the ith layer, ti is the EM

wave two-way travel time through the ith layer as shown
in Fig. 1, c is the speed of light in free space (c
108
m/s), and er,i is the dielectric constant of the ith layer.
Electronically, impulse GPR systems function in the
following manner: A trigger pulse is generated in the
GPR control unit. This trigger pulse is sent to a transceiver, where it is modulated and amplied to become
a bipolar transmit pulse with a much higher amplitude.
The generated pulse is then sent through the transmitting antenna to the ground. After a short time (10100
ns, depending on the antenna used), the reected signal
is collected by the receiving antenna and is transmitted
to the receiver circuitry, where it is ltered and digitized.
Finally, the produced data is displayed for immediate
interpretation and is stored on magnetic media for later
processing.
Depending on the way antennas are used, GPR systems are classied as air-coupled or ground-coupled systems. In air-coupled systems, the antennas (usually horn
antennas) are typically deployed 150500 mm above the
surface. These systems give a clean radar signal and allow for highway speed surveys. However, because part
of the EM energy sent by the antenna is reected back
by the pavement surface, the depth of penetration is limited. In contrast, a ground-coupled systems antenna is
in full contact with the ground, which gives a higher
depth of penetration (at the same frequency) but limits
the speed of the survey. For pavement surveys, GPR
antennae are typically rigidly mounted on a survey
van, as depicted in Fig. 2. This gure shows an air-coupled system composed of a pair of separate horn antennae (bistatic: one serves as a transmitter and the other as
a receiver) and a ground-coupled system comprised of a
single antenna. For exible pavements, the air-coupled
system is usually preferred to the ground-coupled system

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

Fig. 2. Typical GPR van used for pavement surveys showing antennae
conguration.

since HMA layers are not very lossy on the one hand,
and high speed surveys are more economical because
they do not require lane closures, on the other hand.
Moreover, data collected by the air-coupled system allows estimation of the dielectric properties of the layers,
as would be shown in the next section, which are necessary for accurate GPR data interpretation. All GPR
data presented in this paper were collected by an aircoupled GPR system.
Finally, it should be noted that for easier data interpretation, the GPR antennas are usually designed to
radiate a wave that can be approximated in the far eld
[11] by a normal-incidence transverse electromagnetic
(TEM) plane wave. This applies also to the case of a bistatic conguration, where the incidence angle is small
but dierent from zero.

3. Dielectric constant estimation from GPR data

In general, the thickness of any pavement layer can
be estimated based on Eq. (1). Assuming that the twoway travel time ti can be measured accurately from the
GPR reected signal, as pictured in Fig. 1, the dielectric
constant er,i would be the only unknown remaining in
Eq. (1). Assuming that no multiple reections are present within the GPR signal (condition usually exists for
GPR data collected from pavement systems because
layer interfaces are not strong reectors), it can be
shown that the relative reection amplitude at the nth
layer interface is given in to the following equation [12]:
n
#
P
p p "Y
ri d i
p

g0
er;n  er;n1 n1
er;i
An
2
p p
1  ci e i0
;
er;n er;n1 i0
Ainc
n 0; 1; . . . ; N  1;

where N is the number of layers composing the pavement system, An is the reection amplitude at the nth

765

layer interface, er,n and rn are, respectively, the dielectric

constant and conductivity of the nth layer (er,0 = 1 and
r0 = 0 for the air layer underneath the antenna), ci is
the reection coecient at the ith interface given in
Eq. (3), g0 is the wave impedance of free space
(g0
120pX), and Ainc is the amplitude of the incident
GPR signal, which is determined by collecting GPR
data over a large and at copper plate placed on the
pavement surface. Since copper is a good conductor, it
can be considered as a perfect EM reector (with a
reection coecient = 1). Thus, the copper plate reected signal can be assumed to be equal to the reverse
of the incident GPR signal. The copper plate calibration
measurement is usually conducted either at the beginning or at the end of each GPR survey
p p
er;i  er;i1
ci p p .
3
er;i er;i1
The dielectric constant er,1 of the rst (top) layer
(which can be HMA or concrete) can be estimated from
Eq. (2) by substituting n by 0

2
1  A0 =Ainc 
er;1
;
4
1 A0 =Ainc 
where er,1 is the dielectric constant of the top layer and
A0 is the amplitude of the surface reection as shown
in Fig. 1.
The dielectric constant of the second layer er,2 is calculated using Eq. (2) for n equals 1 and using the value
of er,1 computed in Eq. (4)
0
12
r1 t1 c
2 g0 2er;1
1  A0 =Ainc  e
 A1 =Ainc A
er;2 er;1 @
;
5
r1 t1 c
2 g0 2er;1
A1 =Ainc 
1  A0 =Ainc  e
where A1 is the amplitude of the reection obtained at
the interface between the rst and second layers as
shown in Fig. 1, and r1 and t1 are, respectively, the conductivity and two-way travel time of the rst layer.
Similarly, for the third layer the dielectric constant
er,3 is found, according to to the following equation,
by replacing n in Eq. (2) by 2:
0


B1  A0 =Ainc 2 e
er;3 er;2 B
@

1  A0 =Ainc 2 e

g0 c
2

g0 c
2




r1 t1 r2 t2
er;1 er;2
r1 t1 r2 t2
er;1 er;2




12
 c1 A1 =Ainc  A2 =Ainc C
C;
A
 c1 A1 =Ainc A2 =Ainc

6
where A2 is the amplitude of the reection obtained at
the interface between the second and third layers as
shown in Fig. 1. The coecient c1 is the reection coefcient at the interface between the rst and second layers given by Eq. (3) for i equals 1.
The same process can be repeated iteratively to compute the dielectric constant er,n of the nth layer as follows:

766

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

er;n


nP
1
1

A
=A


exp
 g20 c
0
inc
B
B
i1

er;n1 B
nP
1
@
2
1  A0 =Ainc  exp  g20 c
2

i1

ri ti
er;i
ri ti
er;i






nP
2
i1
nP
2

12
ci Ai =Ainc  An1 =Ainc C
C
C.
A
ci Ai =Ainc An1 =Ainc

i1

It should be noted that in the above formulation, the

pavement layers are assumed to be homogeneous.
Therefore, the dielectric constant of each layer is assumed to be constant in the sense that it does not vary
within the layer thickness.

4. GPR detection limitations

As part of an ongoing research project at the Virginia
Smart Road [13], 31 copper plates measuring 914 1219
mm each were embedded in the pavement during construction at the dierent layer interfaces of 12 experimental exible sections. It is important to note that
because of its high conductivity (5.7 107 S/m), copper
is considered a perfect EM reecting material whose
GPR signature could be easily separated from the normal pavement response. Thus, the embedded copper

plates serve as a good indicator of the exact locations

of the pavement interface reections.
To study the GPR detection limitations, researchers
examined the data collected with an air-coupled antenna
from one of the exible sections at the Virginia Smart
Road. The section is composed of the following layers:
wearing surface (WS), HMA base layer (BM-25.0), asphalt-treated open-graded drainage layer (OGDL), cement-stabilized limestone base layer (21-A), limestone
aggregate base layer (21-B), and subgrade layer. Fig.
3(a) shows a scope view of a single GPR scan, and
Fig. 3(b) illustrates a linescan (also known as B-scan)
view of a group of scans collected over the same section.
A linescan view represents a set of along-track scans
stacked together vertically. The amplitude of each scan
is quantized and coded into a solid color. Therefore,
the x-axis in this gure represents the scan number,
which is proportional to the surveyed distance, and the

Fig. 3. (a) Scope view of a GPR scan collected over the Virginia Smart Road, (b) Linescan view showing the copper plate reections at the (1) 21-B/
Subgrade, (2) 21-A/21-B, (3) OGDL/21-A, (4) BM-25.0/OGDL, and (5) WS/BM-25.0 interfaces.

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

y-axis represents the reection time that can be converted to depth, knowing the dielectric properties of
each layer. The amplitude to color transformation function used to obtain Fig. 3(b) is given at the right side of
the gure. In order to enhance low reections, a nonlinear transformation function was used.
As can be seen in Fig. 3(a), only two reected pulses
are visible: the surface reection and the HMA/Base
interface reection. In Fig. 3(b), ve strong reections
can be distinguished. These reections correspond to
the ve copper plates placed at the wearing surface/
BM-25.0, BM-25.0/OGDL, OGDL/21-A, 21-A/21-B,
and 21-B/subgrade interfaces, respectively.
It is clear from Figs. 3(a) and (b) that the 21-B/subgrade and 21-A/21-B interfaces are not detectable under
normal conditions (i.e., without copper plates). This
lack of visibility is due to the low contrast between the
dielectric properties of the subgrade, 21-B, and 21-A layers since they were constructed using the same type of
limestone material (which has a dielectric constant, er,
of approximately 8). The low contrast in dielectric constant results in a low amplitude reected signal, which is
further attenuated by material loss. The eect of material loss on the reected signal can be further shown in
Fig. 3(b), where the reected signal from the deeper copper plates is found to have a lower amplitude (darker
color) than the reected signal from the shallower plates.
In contrast to the base layers, the OGDL/21-A interface
(reection 3 in Fig. 3(b)) is easily detectable, even in the
absence of a copper plate at the interface, as illustrated
in Figs. 3(a) and (b). This is due to the relatively high
contrast between the dielectric constants of HMA (er
approximately 4) and cement-stabilized limestone aggregate (er approximately 8). The high contrast between the
dielectric constants results in a high amplitude reected
signal.
According to the design of the pavement section studied, the HMA layer is composed of a wearing surface, an
HMA base layer (BM-25.0), and an OGDL. However,
in Figs. 3(a) and (b), it is not obvious where the reections from the interfaces between these layers occur. In
fact, a close examination of Fig. 3(b) shows a longitudinal reection within the HMA layers that does not correspond to any copper plate reection depth (the
reection is between reections 4 and 5 in Fig. 3(b),
which correspond to the copper plate reections). This
spurious reection is due to the overlap between the
reections from the pavement surface, wearing surface/BM-25.0 interface, and BM-25.0/OGDL interface.
In this case, the reections within the HMA layer are
masked by the stronger reections in their vicinity (surface reection), rather than by the receiver noise. Consequently, if the three HMA layers are considered as one
layer during the GPR data interpretation, the assumption that the layers are homogeneous would break
down, and the thickness results would be either underes-

767

timated or overestimated, depending on the dierences

between the dielectric constants of the individual layers.

5. Thickness estimation results

5.1. Pavements with thick layers
When the surveyed pavement system is composed of
relatively thick layers, the GPR reected pulses would
have a minor to a non-existent overlap. This condition
makes detecting the layer interface reection easier than
in the case of thin layers. A pavement layer could be
considered as thin or thick depending on whether its
thickness is smaller or larger than the GPR depth resolution Dd, which is given in the following equation [14]
cT
Dd p ;
2 er

where T is the transmitted pulse width and er is the

dielectric constant of the considered layer.
When the pavement layers are relatively thick, it is
guaranteed that the layers detected by GPR are
relatively homogeneous (in the sense that they are not
composed of multiple thin layers), thus reducing thickness-measurement errors caused by dielectric constant
variations within the layer. For rigid pavements, the
condition of thick layers is usually veried for both
the concrete slab and the supporting base layer. In contrast, for exible pavements, the HMA layers are usually
composed of at least one thin layer, which is the wearing
surface or any newly placed overlay. Therefore, the
thick layers condition is generally not applicable for exible pavements. However, in the case of newly built and
non-aged exible pavements, the HMA layers could be
considered as a relatively single homogeneous thick
layer, especially when they are composed of the same
aggregate type.
To validate exible pavement layer thickness estimation using GPR, dierent GPR surveys were conducted
in a quality control-quality assurance (QC/QA) study of
a newly built, three-lane pavement section of Route 288
in Richmond, Virginia [15]. The pavement system was
composed of three HMA layers (HMA base, HMA
intermediate 1, and HMA intermediate 2), with a total
thickness of 240 mm and a 150-mm aggregate base
layer. In order to ensure that the HMA layers were relatively homogeneous for the GPR survey, the GPR
measurements were conducted on the HMA pavement
layers just a few hours after they were placed.
Fig. 4 shows a comparison between the GPR estimated thicknesses, calculated using the aforementioned
data analysis technique, and the design thicknesses for
part of the test section of Route 288. The GPR estimated
thicknesses are very comparable to the design thicknesses. To verify the accuracy of the GPR data analysis,

768

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

Distance (m)
40

45

50

55

60

0
HMA

50

Base

Thickness (mm)

100

HMA Design

150

Base Design

200
250
300
350
400
450

Fig. 4. Comparison between GPR estimated thicknesses and design thicknesses for the HMA and base layers of Route 288.

stationary GPR measurements were collected at specic

locations where some cores were taken for density and
material property tests. The cores were also used to directly measure the HMA layer thicknesses and to compare them to the thicknesses estimated by GPR. Fig. 5
depicts the correlation between the layer thicknesses estimated by GPR and the thicknesses measured directly
from cores for the three HMA layers. This gure shows
that the layer thicknesses found by both techniques are
very comparable (data points congregated around the
line of equity). Actually, the average error between
GPR thickness and core thickness for all three HMA layers was 2.9%. This error is similar to the error found
when measuring layer thicknesses directly on cores [9].
5.2. Thin layer eects on GPR results
Thin layers are typically found on in-service exible
or rigid pavements where thin overlays have been added
over time as part of rehabilitation projects. As mentioned previously, the existence of thin layers within a

pavement system has considerable eects on the layer

thickness results reported by GPR. In fact, the overlap
of the pulses reected from the interfaces of thin layers
makes their detection within the GPR reected signal
cumbersome, as shown in Figs. 3(a) and (b). If these layers are ignored during the GPR data analysis phase (i.e.,
if multiple layers with dierent dielectric properties were
considered as a single homogeneous layer), the dielectric
constants estimated by Eqs. (3) through (7) would be
incorrect. Moreover, since the dielectric constant of
any bottom layer depends on the dielectric constants
of all the layers above it, errors in the estimation of
the top layers dielectric constant would result in erroneous dielectric constants for all the layers underneath.
In order to study the eects of thin layers on GPR
thickness results, GPR scans were collected with the aircoupled antenna over a exible pavement section at the
Virginia Smart Road; these scans were then analyzed to
estimate layer thickness. The total thickness of the considered pavement sections HMA layer was 263 mm, divided
as follows: 38 mm wearing surface (WS), 150 mm HMA

280
HMA Base

GPR Thickness (mm)

260

Avg. Error = 2.16%

HMA Intermediate 1

240

HMA Intermediate 2

220
200
180

Avg. Error = 2.90%

160
140
120

Avg. Error = 3.71%

100
80
80

100

120

140

160
180
200
Core Thickness (mm)

220

240

260

280

Fig. 5. Comparison between GPR estimated HMA thicknesses and core thicknesses.

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

base (BM-25.0), and 75 mm asphalt-treated drainage

layer (OGDL). After construction of the Virginia Smart
Road was completed, direct measurements on cores and
analysis of copper plate location survey data showed that
the as-built HMA thickness diered from the design
thickness by a maximum of 6%. This pavement section
was a good candidate for testing the eects of thin layers
on GPR thickness estimation performance because the
pavement was more than two years old and the layers
had dierent densities thus, resulting in dierent dielectric constants. For example, among the three layers, the
OGDL layer had the highest air-voids and lowest asphalt
content.
As shown in Fig. 3, it is dicult to nd the reection
locations of the individual layers that compose the
HMA layer. This problem is caused, on one hand, by
the overlap between the reected pulses from the WS/
BM-25.0 interface and the pavement surface and, on
the other hand, by the reected pulses from the BM25.0/OGDL interface and the OGDL/Base interface.
Therefore, using the thickness estimation technique
from GPR data would yield the total thickness of
HMA instead of the thicknesses of the individual layers.
The total thickness of the HMA layer found from GPR
data along the test section is presented in Fig. 6. For
comparison purposes, the HMA design thickness for
the same section is also shown in the gure. According
to Fig. 6, GPR results tend to overestimate the real
HMA thickness by approximately 3080 mm, which
represents 1030% of the total design thickness. The
average HMA thickness found for the test section is
315 mm, with a standard deviation of 12 mm or, equivalently, 4%. The average thickness found from GPR
data corresponds to a 20% overestimation when compared to the design thickness. This overestimation is
mainly caused by the following:

20

 The assumption that the three HMA layers have the

same dielectric constant even though they have dierent compositions.
 The dielectric constant computation technique, which
is based on the amplitude of the surface reection, as
given in Eq. (4). In fact, due to the overlap between
the surface reection and the reection at the
WS/BM-25.0 interface, the measured amplitude of
the surface reection would be greater or less than
the real reection amplitude. In other words, due
to the dierent peak polarities of the transmitted
GPR pulses (positive peak surrounded by two negative peaks), the superposition of the two adjacent
reections might have an additive or a subtractive
eect, which will either increase or decrease the measured surface reection amplitude.
In order to increase the accuracy of the GPR depth
estimation technique in the case of thin layers, the average dielectric constant of the three layers should be estimated. One way of determining this average dielectric
constant is to base its computation on the reection
at the OGDL/Base interface, instead of the surface
reection, using a modied common depth technique
proposed by Lahouar et al. [6]. This technique considers the dielectric constant of the whole HMA layer,
rather than the surface area alone. For this solution,
GPR data should be collected simultaneously with
two antenna systems: a bistatic air-coupled antenna
and a monostatic ground-coupled antenna. The dielectric constant estimation is then based on the two-way
travel times of the reections rather than on their
amplitudes.
When GPR data is collected by a single antenna system, an alternative solution to counteract the problem
caused by thin layers would be to resolve the individual

Distance (m)
40
60

80

100

0
50

GPR

Design

100

Depth (mm)

769

150
200
250
300
350
400

Fig. 6. HMA thickness found from GPR data when considering all HMA layers as a single homogeneous layer.

770

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

An alternative deconvolution algorithm that could be

applied for GPR data is to detect the strong reected
pulses in the GPR signal and then use them in a least
squares tting model to accurately estimate the reection amplitudes and generate a theoretical signal containing only strong reections. The synthesized signal
could then be subtracted from the measured GPR signal
to reveal the masked weak reections. The whole process can then be repeated until all the reections present
within the GPR signal are detected and their time delays
and amplitudes are estimated. The detailed algorithm
for this technique is presented in [12].
Fig. 7 depicts the same GPR data as in Fig. 3(b) after
applying the deconvolution algorithm. The gure shows
that the interface reections of the three HMA layers
(WS, BM-25.0, and OGDL) became visible after applying deconvolution. Fig. 8 shows the HMA thicknesses of
the WS, BM-25.0, and OGDL layers found from the
same GPR data as before but after applying deconvolution and then using the aforementioned thickness estimation technique. The HMA layers were assumed

layers in the GPR reected signal using a signal processing technique known as deconvolution. Ideally, deconvolution would remove the eect of the GPR
incident signal from the GPR reected signal, resulting
in the reectivity function of the pavement system. Since
the reectivity function is composed of narrower pulses
than the original reected signal, separating the interface
reections would be easier, even if the considered layers
are thin. After nding the exact reection locations, Eqs.
(3)(7) can be used to iteratively estimate the dielectric
constants of the dierent layers. Then, Eq. (1) can be
used to nd the thicknesses of the individual layers. It
should be noted here that applying deconvolution to
the reected GPR signals usually produces an output
signal with an increased noise level; mainly because
some conditions required by most deconvolution techniques [16] (such as minimum phase and stationarity)
are not veried for the GPR signals [12]. Because a high
noise level generally makes the detection process of the
reected pulses more dicult, the probability of missing
some interfaces in the deconvolved signal might be high.

Fig. 7. GPR data after deconvolution showing the copper plate reections at the (1) 21-B/subgrade, (2) 21-A/21-B, (3) OGDL/21-A, (4) BM-25.0/
OGDL, and (5) WS/BM-25.0 interfaces.

Distance (m)
0

20

40

60

80

100

WS

50

Depth (mm)

100

BM-25.0

150
200
250

OGDL

300
350

WS
OGDL

BM-25.0
Design

400
Fig. 8. HMA thicknesses estimated from GPR data when considering the individual layers.

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

771

Table 1
Comparison between HMA core thicknesses and GPR thicknesses
Core #

Distance (m)

Total HMA thickness (mm)

Error (%)

Core

Overall

Deconvolution

Overall

Deconvolution

A1
A2
A3
A4
B2
D1
D2
E1
E2
F1
F2
F3
G1
G2
H2
J1
J2
K1
K2

45.0
49.9
54.8
60.8
148.5
307.3
312.6
542.9
547.9
608.9
613.9
619.1
723.3
733.3
798.1
979.2
986.4
1091.2
1096.2

282
273
266
268
283
276
265
292
285
211
210
206
195
204
286
286
355
297
297

313
276
259
252
322
315
301
317
302
247
247
239
209
214
302
330
405
376
379

272
276
259
252
266
271
255
298
283
211
211
204
193
202
270
292
367
269
275

10.9
1.1
2.8
5.9
13.9
14.3
13.4
8.6
5.8
16.9
17.4
16.2
7.2
4.9
5.7
15.3
14.2
26.6
27.6

3.7
1.1
2.8
5.9
5.9
1.7
3.9
2.1
0.8
0.2
0.3
0.8
1.0
1.0
5.5
2.0
3.5
9.4
7.4

12.0

3.1

Average absolute error (%)

lossless, therefore, the conductivities ri were considered

null. For comparison purposes, the gure also shows the
design thicknesses for each layer. According to the gure, the thicknesses estimated by GPR are very comparable to the design thicknesses.
To get a better thickness accuracy estimate of the
GPR results, the thicknesses reported by GPR using
the two analysis techniques (i.e., with and without
deconvolution) were compared to the thicknesses measured directly on HMA cores drilled from the road.
The results of the comparison are presented in Table
1. Based on these results, the average absolute thickness
error found when the thin-layer reections were not separated was 12%. This error dropped to 3% when the
reections were separated through deconvolution. This
result shows the improvement made in the thickness estimation accuracy when proper analysis techniques are
used. It should be noted that the lossless assumption
of the HMA layers did not signicantly aect the layer
thickness results because all GPR measurements were
taken under dry pavement conditions and the pavement
layers were not very thick to considerably attenuate the
GPR signals.

6. Summary
This paper presents an overview of the application of
ground penetrating radar (GPR) for pavement thickness
estimation. The basic theoretical steps typically used for
analyzing GPR data and the simplifying assumptions
behind them are explained. The paper also uses eld
data to illustrate the physical limitations that should

be expected when using GPR technology to assess pavement thicknesses. Field data was also used to evaluate
the performance of GPR in estimating layer thicknesses
for dierent cases. In particular, it was shown that when
the surveyed pavement is composed of thick layers
(relative to the incident GPR pulse width), GPR data
yields acceptable thickness results. On the other hand,
if the pavement has at least one thin layer (i.e., the
wearing surface), the GPR thickness accuracy degrades
considerably. In this case, signal processing techniques
should be applied to the GPR reected signal to enhance
its quality prior to estimating the layer thicknesses.

References
[1] Donohue & Associates, Inc. A nondestructive method for
determining the thickness of sound concrete on older pavements.
Final Report HR-250. IA, USA: Iowa Highway Research Board;
1983.
[2] Smith SS, Scullion T. Development of ground penetrating radar
equipment for detecting pavement condition for preventive
maintenance. Final Report Strategic Highway Research Program
SHRP-672. Washington, DC, USA: National Research Council;
1993. p. 177.
[3] Al-Qadi IL. Using microwave measurements to detect moisture in
hot-mix asphalt. J Test Eval 1992;20(1):4550 [JTEVA, ASTM].
[4] Rmeili, E. and Scullion T. Detecting stripping in asphalt concrete
layers using ground penetrating radar. Paper No. 97-0508.
Washington DC: Transportation Research Board; 1997.
[5] Maser KR. Measurement of as-built conditions using ground
penetrating radar. In: Structural materials technology: an NDT
conference; 1996. p. 617.
[6] Lahouar S, Al-Qadi IL, Loulizi A, Trenton CM, Lee DT.
Approach to determining in situ dielectric constant of pavements:
development and implementation at interstate 81. Transportation

772

[7]

[8]

[9]

[10]

I.L. AL-Qadi, S. Lahouar / Construction and Building Materials 19 (2005) 763772

Research Record No. 1806. Assessing and Evaluating Pavements;
2002. p. 817.
Al-Qadi IL, Loulizi A, Elsei M, Lahouar S. The virginia smart
road: the impact of pavement instrumentation on understanding
pavement performance. The Journal of APPT 2004;83:42766.
Al-Qadi IL, Lahouar S. Use of GPR for thickness measurement
and quality control of exible pavements. The Journal of APPT
2004;83:50128.
Al-Qadi IL, Lahouar S, Loulizi A. Successful application of GPR
for quality assurance/quality control of new pavements, Transportation Research Record No. 1861, January 1216; 2003. p. 8697.
Wimsatt AJ, Scullion T, Ragsdale J, Servos S. The use of ground
penetrating radar data in pavement rehabilitation strategy selection and pavement condition assessment. Paper No. 98-0729.
Washington DC: Transportation Research Board; 1998.

[11] Balanis CA. Advanced engineering electromagnetics. New

York: Wiley; 1989.
[12] Lahouar S. Development of data analysis algorithms for interpretation of ground penetrating radar data. Ph.D. dissertation,
Dept. Elect. Eng., Virginia Tech, Blacksburg, VA; 2003.
[13] Al-Qadi IL, Lahouar S, Loulizi A. GPR calibration facility.
In: Uomoto T, editor. Nondestructive testing in civil engineering 2000. Tokyo, Japan: Elsevier Health Sciences; 2000. p.
50918.
[14] Skolnik ML. Introduction to Radar systems. 2nd ed. New
York: Mc-Graw Hill; 1980.
[15] APAC-Virginia, Inc. Route288.com (VA 288), <http://www.route288.com>; 2001 (retrieved 17.07.02).
[16] Dimri V. Deconvolution and inverse theory, application to
geophysical problems. Amsterdam: Elsevier; 1992.