PERGAMON Computers & Geosciences 25 (1999) 141±149

Radar Unix: a complete package for GPR data processing

Gilles Grandjean *, Herve Durand
BRGM, Research Division, BP 6009 Orleans, France
Received 16 October 1997; revised 12 June 1998

Abstract

A complete package for ground penetrating radar data interpretation including data processing, forward modeling
and a case history database consultation is presented. Running on an Unix operating system, its architecture
consists of a graphical user interface generating batch ®les transmitted to a library of processing routines. This
design allows a better software maintenance and the possibility for the user to run processing or modeling batch
®les by itself and di€ered in time. A case history data base is available and consists of an hypertext document which
can be consulted by using a standard HTML browser. All the software speci®cations are presented through a
realistic example. # 1999 Elsevier Science Ltd. All rights reserved.

Code available at http://www.iang.org/CGEditor/index.htm

Keywords: Ground penetrating radar; Graphical user interface; Data processing; Forward modeling; HTML database

1. Introduction to the re¯ection and the refraction optic laws prior to

being recorded by a surface receiver antenna.
Ground penetrating radar (GPR) is frequently used Some applications dealing with objects detection
in geophysical subsurface sounding, as shown by nu- only require simple processing, and most of the time
merous applications in exploration domains, such as raw data are sucient to locate the target in term of a
civil engineering (Mellett, 1995), hydrogeology (Beres re¯ection or a di€raction anomaly. The full potential
and Haeni, 1991), archaeology (Sternberg and McGill, of the method has not been realized and sophisticated
1995) or geology (Grandjean and Gourry, 1996). data processing and rigorous modeling schemes are
Compared to other nondestructive geophysical increasingly demanded for reliable subsurface predic-
methods, GPR provides a quick and easy way to tions (Grandjean et al., 1997). For this reason, BRGM
image the few ®rst meters of the subsurface with a rela- decided to develop a GPR data processing software
tive good resolution. Conceptually identical to seismic based on a novel concept where the three following com-
surveys, especially to single-fold seismic method, GPR ponents are integrated in the same computing platform:
is an electromagnetic technique ranging in the radio data processing, forward modeling and a case history
frequencies: the source is composed by a surface trans- database. The main objectives were ®rst to implement
mitter antenna emitting an electromagnetic pulse cen- high performance processing routines to ensure the
tered on a nominal frequency ranging from 10 MHz to quality of the results, second to decrease as much as
1 GHz; the wave propagates in the ground according possible processing times to allow industrial appli-
cations, and then, to design menus and multiple win-
dows for users convenience. In this paper, we present
* Corresponding author. Tel.: +33-38-643-475; fax: +33- the software design and illustrate the di€erent menu
38-643-361; e-mail: [email protected]. options through the processing of a realistic example.

0098-3004/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 9 8 - 3 0 0 4 ( 9 8 ) 0 0 1 1 3 - 7
142 G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149

2. Software design

In order to use high performance codes without hav-

ing high consuming processing times, we developed the
software on an Unix machine, using an Ansi-C pro-
gramming and a X-Motif based interface. The software
architecture is composed of two main modules: the
graphic user interface (GUI) and the processing routines
library. These two components are compiled separately
and communicate by exchanging batch ®les, allowing the
user to build them with the GUI or by itself (Fig. 1). In
the same way, the processing information generated by
each routines is merged in a result ®le which can be either
directly edited by the user in a text editor, or consulted in
the result window of the GUI.
The maintenance of the processing codes is crucial
for a company, so that any modi®cation of the codes
will not to disturb the daily use of the software. The
programs have to be easily recompiled every time a
computer or an operating system changes, or they
have to be quickly replaced by others if better algor- Fig. 1. Presentation of the architecture of software. User can
ithms are found. For these reasons, the processing rou- build processing or modeling batch ®les either by using
tines only include mathematical operations, they are Graphic User Interface, or by itself with text editor. Unix
written in standard Ansi-C with inputs passed as argu- scripts call processing and modeling routines from library and
ments. Another advantage of this architecture is the are executed. Returned information is written in a log ®le or
possibility for the user to build the processing batch in a GUI result window.
®les with a text editor. This option allows the proces-
remarks about the geological and physical models con-
sing to di€er in time or be run sequentially for multiple
data sets. This architecture is valid for the processing stituting the interpretation. For a quick consultation of
routines including the forward modeling code. the database, we structured it as an hypertext docu-
The case history database references numerous pro- ment. As a consequence, all the texts and GPR images
cessed images, corresponding synthetics and some describing a case history are built in HTML and can

Fig. 2. Look of the GUI main window, showing the menu, the history and the results windows.
G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149 143

Fig. 3. This operation consists of selecting the window which will have to be processed from the raw data. The limits are pointed
by the mouse. The coordinates of this window are indicated to convert each scan in a 3-D reference.

Fig. 4. A Fourier transform can be applied to the scans to evaluate the amplitude spectrum along the scan axis. Here, the fre-
quency bandwidth is restricted from 200 to 600 MHz.
144 G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149

be browsed by using a standard HTML interpreter. dicating if some processing is running or not. The
This technique is also applied to consult the help guide di€erent menu items are described next.
of the software.

3. Menu options 3.1. File

The main GUI window is presented in Fig. 2. It The option File facilitates plotting on the screen and
consists of a menu at the top of the window, a results printing the GPR sections before or after processing.
window on the right side where each returned infor- The size of the plot in centimeters for the horizontal
mation about the processing is displayed, a history and the vertical axes, the labels of axes and the title of
window at the left side where the processing applied to the plot can be speci®ed. The last option is to quit the
data are listed, and a light at the lower right corner in- software.

Fig. 5. This panel shows the di€erent options for the processing routine. Each processing parameters can be stored in a ®le, to be
re called later.
G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149 145

3.2. Editing . Data amplitudes can be transformed into a decibel

scale to get information about the distribution of
This part of the menu is dedicated to importation attenuating zones.
and conversion of data®les from a DZT format, used . A spectral analysis of the section shows the fre-
by the manufacturers of the GPR equipment, to a quency distribution of the data (Fig. 4). This will be
SEGY format more appropriate to the seismic and used afterwards, during the processing, to estimate
GPR data processing algorithms. The following oper- the cu€-o€ values of frequency ®lters.
ations are available: . A velocity ®eld estimation can be performed from a
digitalization of hyperbolic signals using the mouse.
. Horizontal stretching of GPR sections from ®eld This will be useful to have an indication about the
measurement to compensate for the velocity velocity ®eld to correctly migrate the data and pro-
variations of antenna displacements during the duce a more realistic image of the ground.
recording. . Mean attenuation curve can also be plotted to indi-
. The selection of a part of the data. As an example, cate the mean attenuation of the sounded medium.
Fig. 3 shows the screen when a part of the raw data
are selected before processing. GPR data analysis is an important step in the esti-
. The resampling along horizontal distance (scans) mation of ground properties because it governs the
and vertical time (samples) axes. quality of the processing.
. The possibility to make a zoom inside a data win-
dow.

At the end of the editing, data are in the right form 3.4. Processing
to be analyzed and then processed.
The processing procedure consists of a sum of oper-
3.3. Data analysis ations applied to the raw data in order to enhance the
signal-to-noise ratio and to produce a more realistic
This part of the preprocessing gives important indi- image of the ground. Processing routines are standard
cations about the data to perform the more adapted ones selected from seismic methods and adapted to
processing: GPR particularities (Grandjean and Gourry, 1996).

Fig. 6. This screen shows the processed data using the previous parameters. Note that the gray scale shows the envelop of the sig-
nal, i.e. the instantaneous amplitude, which is more representative of the di€raction zones.
146 G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149

Fig. 5 shows the processing panel, where the di€erent . migration: is an operation which transforms the sig-
options are available: nal recorded in the time domain into a new spatially
correct signal distribution (Yilmaz, 1987).
. amplitude compensation: correct for attenuation due
. special ®lters: including powering (Vesnaver and
to spherical divergence, dielectric loss and scattering
Sormani, 1989), instantaneous amplitude conver-
e€ects;
sion and time to depth conversion can also be
. ®lters: four kind of ®lters including mean scan aver-
age, reduction with mean scan average, frequency performed.
and horizontal ®lters are also available to enhance
the signal-to-noise ratio. The ®rst and the second After the di€erent processing operations have been
ones respectively spread and con®ne the signal along selected, the batch ®le can be either saved to be
the distance axis according to a sliding window, the recalled afterwards or run immediately. Fig. 6 shows
third is a band pass ®lter in the frequency domain the result of the processing options selected in Fig. 5.
and the fourth removes the horizontal signal; The speci®c processing applied to data enhance the
. static corrections: remove the topographic e€ects areas, represented in black, where ground heterogene-
according to a compensating velocity value; ities scatter the radar wave.

Fig. 7. This screen shows the build-in editor of the software where the parameter models are described. The language used is expli-
cit and the results can also be stored as a ®le on the disk to be re called later.
G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149 147

Fig. 8. The two parameters, i.e. dielectric constant and factor Q distribution, can be plotted on the screen. Note that random distri-
butions can be used to simulate the scattering e€ects in heterogeneous media.

Fig. 9. Resulting synthetic from the parameter distributions described above.

148 G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149

3.5. Modeling netic wave propagation in heterogeneous medium

(Bitri and Grandjean, 1998). This technique consists of
To produce a reliable interpretation, a GPR section an extrapolation of the electric ®eld in the frequency±
has to be explained by a quantitative model, i.e. a wavenumber domain using a simple multiplication of
physical ground parameters distribution that could the ®eld generated from the sources by the extrapol-
have generated the observed data. For that reason, in ation operator. The method is fast and the grid para-
some case where a high accuracy is needed in the in- meterization allows complex dielectric models to be
terpretation, modeling is an ecient way to verify a tested.
model by calculating synthetic GPR sections. To test The ®rst step of modeling consists of building the
complex distributions of dielectric parameters, we physical model which can be displayed on the screen
developed a modeling method based on electromag- and saved on the disk. Figures 7 and 8 show, respect-

Fig. 10. An example of the HTML french database where all the informations concerning the ®eld survey, the processing modeling
and the interpretation can be consulted by using an HTML interpreter.
G. Grandjean, H. Durand / Computers & Geosciences 25 (1999) 141±149 149

ively the part of the GUI allowing the input of ground dimensions. Considering the expansion of the compu-
heterogeneities and the resulting model sections in ter performances, we plan to implement it on a
term of permittivity and factor Q distribution. The Pentium PC under the Linux operating system.
modeling routine is then called to process the synthetic
GPR section (Fig. 9). This procedure can be iteratively
used until a good match between the observed GPR 5. Software distribution
data and the synthetic section is achieved. Another use
of such a modeling scheme can be used for optimizing Radar Unix can be obtained by contacting the
®eld survey design in desk-top studies, where it may be authors ([email protected]). It is freely available,
necessary to evaluate the applicability of GPR method providing that it is not for commercial use and if every
over some hypothetical geological models before going published material derived from it refers to this article.
to the ®eld.

3.6. Database and help Acknowledgements

The database stores all the features of a study, This work is the BRGM contribution 98018. Some
including the ®eld settings, the raw and processed of the routines we used in our software were derived or
GPR data, the physical model and the corresponding adapted from Seismic Unix and we would like to thank
synthetic and ®nally the comments about the the Colorado School of Mines for this contribution.
interpretation. All case histories can be recalled and
consulted by using an HTML browser. (Fig. 10). This
database is an important step in the GPR data in- References
terpretation, because the experience of every user can
be delivered to others. Providing the database is Beres, M., Jr., Haeni, F.P., 1991. Application of ground pene-
located on the Intranet network of the company, trating radar methods in hydrogeologic studied. Ground
everybody can share these informations and the di€u- Water 29 (3), 375±386.
sion of results can considerably get improved. Bitri, A., Grandjean, G., 1998. Frequency±wavenumber mod-
Help and informations about the use of the software eling and migration of 2-D GPR data in moderately hetero-
are also presented in the form of HTML documents. geneous dispersive media. Geophysical Prospecting 46 (3),
The hypertext facilities of this kind of document pro- 287±301.
Grandjean, G., Bitri, A., Gourry, J.C., 1997. Forward model-
vide a quick and user friendly interface ensuring the
ling contribution in the GPR signatures interpretation of
time to search the required information is minimum. urban heterogeneities. In: EEGS Proceedings, 3th meeting,
Aarhus, Denmark, pp. 75±78.
Grandjean, G., Gourry, J.C., 1996. GPR data processing for
4. Conclusions 3-D fracture mapping in a marble quarry (Thassos,
Greece). Journal of Applied Geophysics 36, 19±30.
The processing software presented in this paper is Mellett, J.S., 1995. Ground penetrating radar applications in
based on a novel concept where data processing, for- engineering, environmental management and geology.
ward modeling and a case history database are inte- Journal of Applied Geophysics 33, 157±166.
grated in the same processing tool. Solutions to Sternberg, B.K., McGill, J.W., 1995. Archaeological studies in
southern Arizona using ground penetrating radar. Journal
implement accuracy algorithms with low processing
of Applied Geophysics 33, 209±225.
times, and to design a friendly interface are proposed Vesnaver, A., Sormani, L., 1989. Frequency domain powering
and ensure that the software can be used for industrial in one and two dimension. Bollettino di Geo®sica Theorica
applications. Actually the system is running on a Sun ed Applicata 31 (121), 3±17.
workstation Sparc 20. The processing and modeling Yilmaz, O., 1987. Seismic data processing. In: Doherty, S.M.
times are from 1 to 5 min depending of the model (Ed.), Series of Investigations in Geophysics, vol. 2, 526 pp.