You are on page 1of 14

Near Surface Geophysics, 2006, 227-240

Analysis of multi-offset GPR data: a case study in a


coarse-grained gravel aquifer
A. Becht1,2*, E. Appel1 and P. Dietrich1,3
1
Institut für Geowissenschaften, Universität Tübingen, Sigwartstraße 10, D-72076 Tübingen, Germany
2
Now at: Statoil ASA, N-4035 Stavanger, Norway
3
Now at: Umweltforschungszentrum, Leipzich-Halle GmbH, Permoserstraße 15, D-04318 Leipzig, Germany

Received August 2004, revision accepted November 2005

ABSTRACT
Multi-offset ground-penetrating radar (GPR) data were collected in a coarse-grained gravel aquifer
located in a glacial delta environment within dipping foresets (at the Tettnang aquifer test site, SW
Germany). We apply prestack processing techniques and normal-moveout (NMO) velocity analysis
in preparation for stacking. The estimation of propagation velocities is of considerable importance
for converting time-domain radargrams into depth-sections and for an interpretation in terms of
petrophysical properties. In our case, interval velocity determination is difficult because reflector
dip angles are variable, and stacking and NMO velocities differ significantly since the antenna
offset is comparable to the reflector depth. Using a synthetic two-layer model, we systematically
study possible errors in interval velocity determination. We compare the cases of decreasing veloc-
ity with depth (typical for ground-penetrating (GPR) surveys of aquifers) and increasing velocity
with depth (typical for seismic surveys). In the case of decreasing velocity with depth, the discrep-
ancy between stacking and NMO velocity is considerable and, consequently, interval velocities
calculated with the Dix equation or the more accurate 2D NMO approximation for dipping layers
show unacceptably high errors. The errors are much smaller in comparable models with increasing
velocity with depth. Thus, the small-offset approximation inherent in the NMO concept is not
appropriate for the first case. Consequently, it is necessary to model the true common-midpoint
(CMP) raypaths to determine realistic interval velocities from our GPR field data.
Stacked GPR sections of multi-offset data show significant improvements in imaging of reflec-
tors and depth of investigation compared with a standard constant-offset section. This allows a more
reliable interpretation of geological structures. Simultaneous acquisition of four channels, which is
possible with some commercial GPR systems, yields significantly better results than constant-offset
standard acquisition without increasing the efforts for data acquisition. Major reflectors in the GPR
section can be correlated with distinct contrasts of porosity, represented by the base of open-frame-
work gravels and sand beds within poorly sorted gravels.

INTRODUCTION GPR data are most commonly acquired along single lines
Accurate investigation of the subsurface structure is required in using a constant-offset transmitter–receiver configuration.
many environmental and engineering applications and sedimen- Data acquisition and 2D processing are relatively simple and
tological studies. The conventional method of drilling yields only fast, allowing a rapid interpretation of GPR lines. In the past
sparse 1D information, which is often insufficient for an accurate decade, research has been focused on GPR acquisition and
site characterization. Therefore, 2D or 3D geophysical techniques processing techniques for improved imaging and interpretation
are increasingly used for high-resolution imaging of subsurface of subsurface structures. Certain processing steps used for
structures. Ground-penetrating radar (GPR) is a promising tech- correctly imaging the geometry of subsurface reflectors (e.g.
nique for high-resolution investigations of the shallow subsurface migration, time-to-depth conversion) need reliable velocity mod-
and has been used in a variety of applications for the last two els. Frequently, multi-offset common-midpoint (CMP) measure-
decades. The GPR method is best applicable in low-loss media ments are used to determine subsurface velocities. Recent
with low silt and clay content and fresh pore-water. studies have shown that, when using the CMP acquisition geom-
etry for entire profiles, multitrace seismic processing techniques
*abecht@gmx.de can be applied to enhance GPR images of the subsurface (Fisher

© 2006 European Association of Geoscientists & Engineers 227


228 A. Becht, E. Appel and P. Dietrich

et al. 1992; Greaves et al. 1996; Hollender et al. 1999; Pipan et


al. 1999). Prospects of multi-offset data acquisition and process-
ing include reducing random and correlated noise, estimating 2D
or 3D velocity models, and applying advanced processing
techniques, such as amplitude-versus-offset analysis. In addition
to improving GPR images, velocity models based on multi-offset
GPR profiling were, for example, used to estimate the water-
content variations of the subsurface (Greaves et al. 1996;
Garambois et al. 2002). However, multi-offset GPR is still not
widely used, because CMP data acquisition for entire profiles is
very time-consuming. More efficient multi-offset data collection
is nowadays possible through advances in commercial
multichannel GPR systems, which allow simultaneous recording
of up to four or even 16 channels with different offsets.
In this study, we describe the acquisition, processing and inter-
pretation of multi-offset GPR data at a coarse-grained gravel aqui-
fer in SW Germany. The test site is situated in foreset deposits of a
glacial delta environment and thus encompasses dipping gravel
beds with highly contrasting hydraulic properties. Special emphasis
is placed on the velocity analysis of CMP data, because reliable
velocity information is of considerable importance for converting
traveltime-domain radargrams into depth-domain geological sec-
tions and for possible interpretation in terms of petrophysical
parameters. Potential errors in interval velocity determination FIGURE 1
related to simplifications inherent in the normal-moveout (NMO) Location map of the test site.
concept are discussed with a synthetic modelling study. We com-
pare stacked sections of different fold orders (maximum- and four-
fold stacked) with a constant-offset section representing a standard
GPR survey and show improvements in the stacked multi-offset
data. The maximum-stacked GPR section is interpreted geologi-
cally with the aid of drillings located along the GPR line.

STUDY AREA
The study area is located in a gravel pit close to Lake Constance in
SW Germany (see Fig. 1). The sediments encompass gravel beds
of a coarse-grained delta deposited during the last Alpine glaciation
period. We established a test site within the dipping foresets of the
delta, which generally show dip angles of 20–30°. Five boreholes
were drilled to a depth of 25 m for detailed geophysical and hydrau-
lic investigations in order to generate a large database for compari-
sons and evaluations of the methods. The groundwater level was FIGURE 2
about 3 m below the ground surface, depending on seasonal varia- Outcrop photograph showing dipping foresets of the coarse-grained
tions. The sedimentology was studied in adjoining outcrops and gravel delta.
could be correlated with the subsurface at the test site. An outcrop
photograph near the test site is shown in Fig. 2. fer. The deposits show extremely contrasting hydraulic proper-
The drilling cores were described in detail, based on a litho- ties; Table 1 gives an overview of the typical gravel types at the
facies type classification scheme developed from outcrop studies test site. More details about the lithofacies types and hydraulic
(Kostic 2004). Differentiation of lithofacies types on the delta properties have been given by Kostic (2004).
slope is controlled by gravitational mass transport processes.
Typical lithofacies types are poorly sorted gravels, bimodal grav- DATA ACQUISITION
els (cobbles in a sandy matrix), open-framework gravels (frame- We acquired multi-offset GPR data during two days in May 2003.
work of cobbles free of matrix), and sand or gravel-sand beds. The GPR line was parallel to the drillings B1–B3, orientated in
The lithofacies types control the hydraulic properties of the aqui- the dip direction of the gravel foresets (see Fig. 3). The ground

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
Analysis of multi-offset GPR data 229

TABLE 1
Overview of average porosities and hydraulic conductivities for the characteristic gravel types at the test site (modified after Kostic 2004). The abbre-
viation are as follows: G – gravel, S – sand, b – bimodal, c – cobble-rich, o – open-framework; e.g. cG,o indicates cobble-rich gravel, open-framework.
Porosity data are average values from laboratory measurements based on numerous samples from various locations (Heinz et al. 2003). Hydraulic
conductivities are geometric means of measured values (based on flowmeter field data) and calculated values (based on grain-size distributions using
the Kozeny–Carman equation and Beyer (1964) method) at the described test site.

Gravel type Porosity Hydraulic conductivity (m/s)

Open-framework gravel (cG,o) 0.26 1.9 ×10-1


Bimodal gravel (cG,b) 0.20 2.2 ×10-4
Poorly sorted gravel (G; cG) 0.18 1.2 ×10-3
Gravel-sand (GS) 0.27 2.1 ×10-4
Sand (S) 0.36 1.0 ×10-4

surface along the multi-offset GPR line was horizontal without and post-stack processing (migration, time-to-depth conver-
significant topography. The groundwater level was at 3.2 m sion). Each of these processing steps is discussed in turn.
below the ground surface during data acquisition. We give a comprehensive discussion of the subject of veloc-
We used a GSSI SIR 2000 system (Geophysical Survey ity analysis of CMP data, because it is the crucial point for
Systems Inc., North Salem, NH, USA) and shielded antennae converting time-domain radargram data to depth-domain
with a nominal frequency of 100 MHz. We applied the common- geological sections and for interpreting velocity information in
receiver-point (CRP) mode for data acquisition and used a CRP terms of petrophysical properties. For this, we briefly review the
spacing of 0.25 m. Transmitter offsets ranged from 1 to 10 m NMO concept and discuss possible errors in interval velocity
with an offset increment of 0.25 m. In total, we collected 97 CRP determination with a synthetic modelling study.
data sets with 37 traces per CRP. After re-sorting the data from
CRP to CMP gathers, a maximum stacking coverage of 18 traces Prestack processing
per CMP can be achieved by this acquisition geometry. To The multi-offset GPR data were processed using the commer-
enhance the signal-to-noise ratio, we used 32 vertical stacks per cially available software package ReflexW (Sandmeier Scientific
trace. The total record time and number of samples per trace Software, Karlsruhe, Germany). Figure 4(a) shows some raw
were 500 ns and 1024, respectively. Omitting the low-covered data examples of CRP gathers after filtering with a running mean
ends of the line, the total length of the GPR profile after stacking dewow (window length: 14 ns) and applying a static time-zero
was about 20 m. correction. Some hyperbolic arrivals from reflections can be
identified in these raw data; however, system ringing (horizontal
DATA PROCESSING arrivals) is significant and the signal-to-noise ratio is therefore
The data processing of multi-offset GPR data follows the classi- relatively low. System ringing is a well-known artefact, which
cal scheme of multi-channel reflection seismics, which is well may be caused by feedback effects within the GPR system. To
documented in the literature (e.g. Hatton et al. 1986; Yilmaz reduce coherent and random noise, we processed the data in the
1987). The processing sequence consists of prestack processing following manner.
(amplitude scaling, minimization of coherent and incoherent
noise), velocity analysis, muting, NMO corrections, stacking,

FIGURE 3 FIGURE 4
Photograph of data acquisition site and plan view. Drillings are indicated Examples of CRP gathers: (a) Raw data with AGC (100 ns window) for
by B1–B5. display; (b) data after prestack processing.

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
230 A. Becht, E. Appel and P. Dietrich

We first tried to minimize the effects of system ringing. where x is the source–receiver offset, t0 is the zero-offset time of
System ringing showed up as horizontal events with nearly the normal raypath to the reflector, and vNMO is the NMO
constant amplitudes within individual CRP gathers. We effec- velocity. Solving (1) for the delay time ∆tNMO, or normal
tively minimized this horizontal coherent noise by applying a moveout, of a reflection for a raypath with a source–receiver
sliding background removal filter for each CRP gather individu- offset x compared with the zero-offset normal raypath, we derive
ally. The filter acts as a horizontal high-pass filter and calculates the well-known NMO equation,
an average trace over a window of traces around a mid-trace and
t  t0 .
12
subtracts it from the mid-trace. We used a window width of 't NMO 2
0  x 2 vNMO
2
(2)
seven traces for averaging.
Amplitude scaling was based on amplitude envelopes follow- In the general case of arbitrarily dipping layers, the NMO veloc-
ing an approach described by Gross et al. (2003). This approach ity is approximated by
has advantages over the commonly applied automatic-gain con-
1 N i 1
§ cos 2 D k ·
trol (AGC). After application of an AGC, broad low-amplitude 2
vNMO
t0 cos 2 E 0
¦ v 't – ¨ cos
i 1
2
i i
k 1©
2
Ek ¹
¸, (3)
shadows typically appear after the strong direct-wave first arriv-
als. This effect is reduced with the amplitude-envelope approach. where t0 is the two-way traveltime of the normal ray to the Nth
First, the Hilbert transform is calculated for each trace. The reflector, ∆ti is the two-way traveltime of the normal ray to the
resulting amplitude envelope is normalized on the maximum Nth reflector within the ith layer, and vi is the interval velocity
value and smoothed with a running average filter to reduce the within the ith layer. The angles i and i are the angles of inci-
effects of local events and noise. A threshold value is set to con- dence and refraction of the normal ray at the ith interface, and 0
trol the maximum gain-factor in order not to over-amplify noise. is the angle between the emerging normal ray and the vertical at
After parameter testing, we chose a maximum gain-factor of the surface. Assuming the reflectors are plane surfaces and inter-
2000. Finally, the original traces are divided by their filtered val velocities are constant, the emergence angle 0 of the normal
amplitude envelopes to obtain amplitude-balanced traces. ray is related to the time slope Sx = dt0(x)/dx of a reflector by
The final step of prestack processing was band-pass filtering
with a passband of 30–160 MHz to reduce high-frequency noise. Sx 2sinE 0 v1 , (4)
Figure 4(b) shows the example CRP gathers after applying these
processing steps. The CRP gathers were subsequently re-sorted where v1 is the interval velocity of the first layer. The time slope
into CMP gathers for velocity analysis and stacking. Sx of a reflector can be measured on a zero-offset stacked section.
Using (1), (3), (4) and Snell’s law,
Velocity analysis
In principle, velocity information can be extracted from CMP sin D i vi sin Ei vi 1 , (5)
data by analysing the propagation velocity of the direct ground
wave and by NMO-velocity analysis of reflection hyperbolae. which describes refraction at the ith interface, it is possible to
The latter is important for defining a velocity model for stacking. determine interval velocities, depths and dips of all layers with a
In addition, stacking velocities can be used to estimate interval knowledge of t0, Sx and vNMO for each primary reflection, pro-
velocities representing the average wave-propagation velocities vided the offset is small compared with reflector depth. The
within individual layers of the subsurface. Estimating interval- derivations of these equations and numerical examples have
velocity models is an important and critical point necessary for been given by Shah (1973) for the 2D case and by Hubral (1976)
processing steps, such as migration and time-to-depth conver- for the 3D case.
sion. In the following, we summarize the NMO concept and In the special case of plane horizontal layers and if the
discuss inherent simplifications. source–receiver offset is small compared with the reflector
depth, the NMO velocity can be approximated by the root-mean-
Review of the NMO concept square (rms) velocity, defined by
Let us assume a 2D subsurface geometry with a number of plane
1 N 2
layers with arbitrary dip angles. The wave-propagation velocity vN2 ¦ vi 'ti ,
t0 i 1
(6)
within each layer, or the interval velocity, is assumed to be iso-
tropic and homogeneous. For a CMP arrangement of source and where vN is the rms velocity between the surface and the Nth
receiver positions and if the source–receiver offset is small com- reflector, t0 is the vertical two-way traveltime from the surface to
pared with the reflector depth (small-offset approximation), the the Nth reflector, and ∆ti is the vertical two-way traveltime
time–distance curve is given by the hyperbolic expression, within the ith layer. Assuming we have estimated rms velocities
from NMO velocities, we can calculate interval velocities using
t 2 x t02  x 2 vNMO
2
, (1) the Dix recursive equation (Dix 1955),

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
Analysis of multi-offset GPR data 231

vN2 t N  vN2 1t N 1 (7) contrasts and dipping reflectors. When simplified, we expect a
vi2 ,
t N  t N 1 two-layer velocity model with a fast top layer and a slow bottom
where vN and tN are the rms velocity and the two-way traveltime layer, corresponding to the unsaturated and saturated zones,
from the surface to the Nth reflector, respectively. Basically, the respectively. The velocity within these zones is variable due to
rms velocity is a weighted average of the interval velocities the different porosities of the dipping gravel beds. Determining
above a reflector. In the special case of plane horizontal layers, interval velocities of the saturated zone is difficult because we
the NMO velocity is equal to the rms velocity. In the case of dip- must analyse reflections from variably dipping gravel beds
ping layers, the two quantities may differ significantly. below the horizontal groundwater level.
In the special case of a single dipping reflector or in a multi- Thus, we are faced with the following questions: What is the
layer case with parallel dipping reflectors, the zero-offset raypath magnitude of the discrepancy between stacking velocity and
is straight and normal to all reflectors, and the emergence angle NMO velocity? Consequently, is the small-offset approximation
0
of the normal ray is equal to the dip angle. From (3) and (4), appropriate under such conditions? How large is the error in the
the NMO velocity is approximated by interval velocity estimation based on stacking velocities, when
using the Dix equation or the NMO approximation for dipping
vNMO vN / cos M , (8) layers? Are there differences in the case of decreasing velocity
with depth, which is typical for GPR surveys in aquifer settings,
where is the reflector dip angle. Thus, the NMO velocity of a compared to the case of increasing velocity with depth, which we
dipping reflector increases with the dip angle and overestimates usually encounter in reflection seismic surveys? To discuss these
the rms velocity. This relationship is important when estimating issues, we present a modelling study in the next section.
rms velocities from NMO velocities.
There are a number of different methods for NMO-velocity Modelling study
analysis of CMP data (e.g. Hatton et al. 1986; Yilmaz 1987). To study the effects of large offsets and dipping reflectors on NMO-
A frequently used technique is the method of constant-velocity velocity analysis and interval-velocity determination, we calculated
stacks, where the CMP data are NMO-corrected with a range of synthetic CMP traveltimes for a two-layer case consisting of a dip-
constant velocities. In the processed scans, the velocity at which ping reflector underlying a horizontal reflector as shown in Fig. 5.
the reflector appears most accurately horizontal corresponds to For each CMP transmitter–receiver pair, we forward-calculated the
the correct stacking velocity. Another common technique is sem- raypath for a reflection at the dipping interface using Snell’s law.
blance-amplitude analysis, where CMP data are NMO-corrected Knowing the lengths of the raypath segments in each layer and the
and stacked with a range of trial velocities. In the resultant veloc- layer velocity, we then calculated the traveltime. The geometry of
ity spectrum, peaks represent the optimum stacking velocity.
These methods aim at determining stacking velocities for the
optimum quality of the stacked signal, assuming a hyperbolic
traveltime–offset function as defined in (1).
At this point, we should make a clear distinction between
NMO velocity and stacking velocity, which are often used syn-
onymously in practice. Stacking velocity is an average moveout
velocity which gives optimum stacking results. In contrast,
NMO velocity is the small-offset moveout velocity measured
near the zero-offset normal raypath. If the offset is small com-
pared with reflector depth, the stacking velocity may serve as a
good approximation for NMO velocity. However, if the offset is
large and the velocity field is complicated with, for example,
lateral velocity gradients or steeply dipping layers, NMO can
deviate strongly from hyperbolic and the determined stacking
velocities may differ significantly from NMO velocities.
In multi-offset GPR, the offset range is typically of the same
magnitude as the reflector depth, due to the limited depth of FIGURE 5
investigation and the necessity for large offsets for reliable Model 1 showing the situation of the GPR field survey with shallow
velocity estimations. In fact, constant-offset GPR data collected reflector case (left) and deep reflector case (right). The dip angle of the
at the field site used in this study show investigation depths of second interface is = 20°. (a) Raypaths for second reflector; for clarity,
maximum 10–15 m, whereas the antenna offsets of the multi- every second ray is shown. (b) Modelled traveltimes (crosses), stacking
offset data range up to 10 m. In addition, we may expect a rela- hyperbola (solid line) and NMO hyperbola (dashed line). The model
tively complicated situation at our field site with strong velocity parameters and calculated velocities are listed in Table 2.

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
232 A. Becht, E. Appel and P. Dietrich

the CMP arrangement corresponds to the field data with offsets t1(0) of 50 ns, which corresponds to a thickness of 3.0 m. The
ranging from 0 to 10 m and an offset increment of 0.5 m. second layer represents the saturated zone with a constant veloc-
From the modelled traveltime–offset curve, we determined ity v2 of 0.07 m/ns. The dip angle of the second interface is
the stacking velocity and the NMO velocity. Determination of equal to 20°. We examined two cases: when the dipping reflector
the NMO velocity is based on (1) and consists of fitting a least- is relatively shallow and when it is deep, with corresponding
squares hyperbola to the small-offset traveltimes, where t(0) is zero-offset times t2(0) of 160 ns and 260 ns. The model param-
given by the zero-offset time of the normal raypath. We used the eters and velocity values determined are listed in Table 2.
first three offsets (x = 0–1 m) for fitting. The stacking velocity Figure 5 shows the raypaths and calculated traveltimes together
corresponds to a least-squares hyperbola fitted to the full offset with the stacking and NMO hyperbolae. The modelled travel-
range. Thus, the zero-offset times of the stacking and NMO time–offset curves are not exactly hyperbolic and deviate sig-
hyperbolae may differ slightly. nificantly from the NMO hyperbolae. The stacking velocity
We used two approaches to determine interval velocities from v2,STK overestimates the NMO velocity v2,NMO by 9.7% for the
stacking velocities. The first and simple approach is the Dix shallow case and by 5.5% for the deep reflector case.
equation (7), which assumes horizontal layering and small off- Use of the Dix equation to recover the interval velocity of the
sets compared to reflector depth. The second approach is based second layer results in unacceptably high values, overestimating
on the 2D NMO-velocity approximation for arbitrarily dipping the true interval velocity by 71% and 40% for the shallow and
layers given by (3), which considers dipping layers and also deep cases, respectively. The 2D NMO approach, which consid-
assumes small offsets. The first approach applying the Dix for- ers the correct time slope of the reflector, leads to interval
mula is straightforward and requires the additional assumption velocities that overestimate the true values by 26% and 10% for
that the rms velocity to the second interface is equal to the NMO the shallow and deep cases, respectively. If we used v2,NMO
velocity. For the second approach, we performed the following instead of v2,STK for this second calculation, we would of course
procedure: Calculate the time slope Sx = 2 sin / v2 from (4) and recover the true interval velocity of 0.07 m/ns. Thus, discrepan-
(5) with 0 = 1 and 1 = , where is the dip angle of the second cies in v2,NMO and v2,STK are strongly amplified in this model when
interface. Determine the emergence angle 0 of the normal ray we calculate the interval velocity from the stacking velocity,
from v1 and Sx using (4), and the refraction angle 1 using Snell’s based on the assumption that the stacking velocity approximates
law with 0 = 1. Calculation of the lengths of the normal raypath the NMO velocity sufficiently well. In this example, neither the
segments using t1(0) and t2(0) provides ∆t1 and ∆t2. With these Dix equation nor the 2D NMO approximation for dipping layers
parameters and the assumption that stacking velocity equals NMO is adequate to determine interval velocities from stacking veloc-
velocity, equation (3) can be solved for the interval velocity. ities. The small-offset approximation is not appropriate.
Model 1 resembles the situation of the multi-offset GPR field Figure 6 shows the relative difference between stacking and
survey. The first layer represents the unsaturated zone with a NMO velocities of the dipping interface, computed for dip angles
constant velocity v1 of 0.12 m/ns and a vertical zero-offset time varying between 0° and 30° and for the constant zero-offset times

TABLE 2
Model parameters and calculated velocities. The velocities v2,STK and v2,NMO are the stacking and NMO velocities of reflections from the second
interface. Using v2,STK, the interval velocity v2,DIX is calculated with the Dix formula, v2,2D with the 2D NMO approximation for dipping layers

Parameter Mod. 1, shallow Mod. 1, deep Mod. 2, shallow Mod. 2, deep

Model definition
t1(0) (ns) 50 50 85 85
t2(0) (ns) 160 260 155 210
v1 (m/ns) 0.12 0.12 0.07 0.07
v2 (m/ns) 0.07 0.07 0.12 0.12
(°) 20 20 20 20

Velocity analysis
v2,STK (m/ns) 0.1197 0.1025 0.1015 0.1085
v2,NMO (m/ns) 0.1091 0.0972 0.1002 0.1081

Interval velocities using v2,STK


v2,DIX (m/ns) 0.1196 0.0979 0.1298 0.1283
v2,2D (m/ns) 0.0883 0.0772 0.1218 0.1206

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
Analysis of multi-offset GPR data 233

FIGURE 7
FIGURE 6 Model 2 showing shallow reflector case (left) and deep reflector case
Relative difference between stacking velocity and NMO velocity versus (right). The dip angle of the second interface is = 20°. (a) Raypaths for
dip angle for Model 1 (see Table 2). second reflector; for clarity, every second ray is shown. (b) Modelled
traveltimes (crosses), stacking hyperbola (solid line) and NMO hyper-
t2(0) of 160 and 260 ns. The relative difference between the two bola (dashed line). The model parameters and calculated velocities are
velocities increases non-linearly with increasing dip angle and listed in Table 2.
decreasing zero-offset time or, respectively, reflector depth. This
figure implies that, in the case of Model 1, interval velocities deter-
mined from stacking velocities are very sensitive to the dip angle
or, respectively, the time slope being measured on a zero-offset
stacked section. Note that when = 0° (the horizontal two-layer
case), the relative difference is still about 1–2%. This means that,
even in the horizontal layer case, interval velocities are overesti-
mated when we use the stacking velocity as an approximation for
the NMO velocity and calculate interval velocities with the Dix
equation.
Model 2 is similar to Model 1 but the values of v1 and v2 are
interchanged and hence v1 = 0.07 m/ns and v2 = 0.12 m/ns. The
zero-offset times are chosen such that the vertical depths of the two
interfaces below the CMP are equivalent to Model 1. This case
reflects a situation of increasing velocity with depth which is typi-
cal, for example, in seismic surveys. Thus, this case may serve as a
good example of a shallow reflection seismic experiment, where
the time units and absolute values are, of course, different. In con-
trast to Model 1, the stacking and NMO hyperbolae are almost FIGURE 8
congruent and v2,STK overestimates v2,NMO by only 1.3% for the shal- Relative difference between stacking velocity and NMO velocity versus
low reflector case and 0.44% for the deep reflector case (Fig. 7). dip angle for Model 2 (see Table 2).
Interval velocities calculated with the Dix formula overestimate the
true value by 8.2% and 6.9% for the shallow and deep reflector
cases, respectively. Using the 2D NMO approximation for dipping difference does not increase significantly with dip angle; it depends
reflectors, the corresponding values are only 1.5% and 0.50%. mainly on the zero-offset time or depth of the reflector.
Thus, the errors in interval-velocity determination are much small- Thus, there are considerable differences between the cases of
er compared to Model 1; the 2D NMO approximation yields results decreasing and increasing velocity with depth as a consequence
with very small errors in this case. Figure 8 shows the relative dif- of different raypath geometries. When velocity increases with
ference between stacking velocity and NMO velocity, calculated depth as in Model 2, the errors in interval-velocity calculation
for varying dip angles. In contrast to Model 1 (Fig. 6), the relative are acceptable with the Dix equation and very small with the

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
234 A. Becht, E. Appel and P. Dietrich

more accurate 2D NMO approximation. However, both methods


lead to unacceptable errors when velocity decreases with depth
as in Model 1. For this case, the small-offset approximation is
certainly not appropriate. A more accurate approach that consid-
ers the actual CMP geometry and allows forward calculation of
stacking velocities is necessary for interval velocity estima-
tions.

Velocity analysis of field data


In principle, velocity information can be obtained from CMP
data by analysing the velocities of the direct ground-wave arriv-
als and by NMO-velocity analysis of reflection hyperbolae.
A priori, we may expect a velocity model with at least two layers
consisting of an unsaturated zone with higher velocities and a FIGURE 9
saturated zone with lower velocities. A more detailed differentia- Example of velocity analysis. (a) CMP No. 141 at 18-m profile distance.
tion of the model depends on the presence of consistent and (b) Velocity spectrum. The triangles indicate the stacking velocities of
strong reflectors which can be analysed. In general, the number the analysed reflectors. (c) 1D stacking-velocity model.
of CMPs to be analysed depends on the lateral variability of the
velocity and the geometry of subsurface structures. Ideally, each
CMP could be analysed to determine the optimum stacking-
velocity model for maximum coherence of reflected arrivals.
However, in practice, the resulting 2D stacking-velocity model
for the complete section would tend to be very heterogeneous on
the scale of the CMP spacing, leading to artefacts in the final
stacked section. Therefore, a compromise between lateral
smoothness and resolution of the 2D stacking-velocity model
must be achieved. We performed NMO-velocity analyses at
regular intervals of 2 m along the line. We determined stacking FIGURE 10
velocities from CMP data using semblance-amplitude analysis. Velocity profiles for the unsaturated zone. The solid line represents the
In the resultant velocity spectrum, peaks representing the opti- velocity of the direct ground wave determined for each CMP (spacing =
mum stacking velocity were picked to determine a 1D velocity 0.125 m). The dashed line corresponds to the stacking velocity of reflec-
model for that CMP. In general, the resolution of the velocity tions from the groundwater level, analysed every 0.5 m.
spectrum improves with an increasing number of traces and a
larger offset range of a CMP. We estimate an uncertainty of about
±0.005 m/ns when velocities are picked in single-velocity spec-
tra. As an example, Fig. 9 shows a CMP gather with the calcu-
lated velocity spectrum and determined stacking velocities for
three major reflectors.
In most CMP gathers, the groundwater level is represented by
strong and consistent arrivals at about 53 ns two-way traveltime.
Since it is not possible to determine stacking velocities of reflec-
tions from above the groundwater level, the groundwater level is
the first major reflector to be analysed. The stacking velocities of
these arrivals represent the true average velocities of the unsatu-
rated zone, because the groundwater level is quasi-horizontal
within the survey area. In some places, however, interferences of
reflections from the dipping gravel beds with those from the
groundwater level may result in uncertainties of velocity deter- FIGURE 11
mination. Figure 10 shows velocity profiles for the unsaturated Stacking-velocity model interpolated from 1D velocity functions. The
zone based on NMO-velocity analysis of reflections from the numbers indicate stacking velocities in m/ns. The thick lines represent
groundwater level and on velocities of the direct ground wave. the reflectors for which stacking velocities were determined. Velocities
The latter were determined by regression analysis of the arrival below the bottom reflector were extrapolated for the purpose of NMO
times of the first positive phase of the direct ground wave for correction.

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
Analysis of multi-offset GPR data 235

offsets from 2 m to 6 m. Both velocity profiles show the same


mean value of 0.122 m/ns and relatively small velocity variations
along the profile without significant trends. Discrepancies
between the two profiles (e.g. at 11 m) may be caused by subsur-
face heterogeneity and different sampling volumes of the direct
and reflected waves used for these analyses. Due to these
discrepancies and the relatively small variations in velocity
values, we assigned a constant velocity of 0.122 m/ns (mean
value) and a constant thickness of 3.2 m (depth of groundwater
level) to the first layer of the stacking-velocity model.
Strong and consistent reflections from below the groundwater
level were not present in all CMPs. Therefore, stacking veloci-
ties for the saturated zone could only be determined for two FIGURE 12
major reflectors. If reflections were too weak or incoherent, we (a) Interval-velocity model determined from inversion of stacking
interpolated the velocity models from neighbouring CMPs. velocities of the bottom major reflector (compare with Fig. 11). Velocities
Figure 11 shows the 2D stacking-velocity model created by below that reflector cannot be determined from the stacking-velocity
interpolating the 1D velocity functions at the analysed CMP model. (b) Tomographic velocity reconstruction between drillings B2
locations. Velocities below the lowermost reflector are extrapo- and B3. The thick line indicates the bottom reflector; the numbers
lated for the purpose of NMO correction. As was shown in the indicate velocities in m/ns.
modelling study, stacking velocities are very sensitive to the
reflector dip angle and obviously vary with traveltime. Therefore, TABLE 3
structures in the stacking velocity model should not be inter- Uncertainty in interval-velocity calculation. Calculated interval veloci-
preted in terms of geological structures or trends. The computa- ties of the saturated zone in m/ns for time slopes sx and stacking veloci-
tion of interval velocities is the topic of the next section. ties v2,STK, varying within the limits of estimated parameter uncertainties.
The model parameters correspond to the CMP at 12-m profile distance.
Interval velocities Bold values indicate the actually determined values for that CMP.
Determining interval velocities from the stacking-velocity model
is complicated by the facts that the NMO small-offset approxi- Sx (ns/m)
mation is not appropriate and the geometry consists of a dipping
reflector underlying a horizontal interface. As shown in the mod- 7.6 8.1 8.6
elling study, use of the Dix equation would lead to enormous
errors in interval-velocity estimates, of up to 70% in this specific 0.101 0.065 0.055 0.037
case. Using the NMO approximation for dipping layers would
v2,STK (m/ns)

also lead to significant errors of the order of 10–25%, depending 0.106 0.078 0.072 0.063
on reflector depth. Therefore, we use a ray-modelling approach
for inversion of stacking velocities. For an initial interval-veloc- 0.111 0.088 0.084 0.078
ity model and a given CMP geometry, the stacking velocity is
forward-calculated. The correct interval velocity is determined
by minimizing the discrepancy between the stacking velocity of 0.072 m/ns. For comparison, Fig. 12(b) shows the upper part of a
the field data and the modelled value. The model layout is equal cross-hole tomographic velocity reconstruction between boreholes
to that of the modelling study described above. The parameters B2 and B3. For this tomographic survey, we used a Ramac bore-
of the horizontal first layer were fixed at v1 = 0.122 m/ns (mean hole radar system (Malå Geoscience, Sweden) and antennae with
velocity of unsaturated zone) and t1(0) = 53 ns (mean traveltime a nominal frequency of 250 MHz. The velocities of the CMP-
to the groundwater level). We used the bottom reflector to deter- derived interval-velocity model are consistently higher but gener-
mine average interval velocities for the saturated zone (see ally agree with the tomographic velocity reconstruction. The bot-
Fig. 11). For each analysed CMP location, we measured the time tom reflector is indicated by a thick line in the tomographic result
slope Sx and zero-offset time t2(0) of this reflector from a stacked and corresponds to a sharp velocity increase.
section. We then determined the correct interval velocity of the It is well known in seismic surveying that interval-velocity
second layer, thus providing 1D interval-velocity functions at the calculations tend to be unstable and are very sensitive to the
analysed CMP locations. accuracy of NMO velocities and other parameters (Hajnal and
Figure 12(a) shows the resultant interval velocity model Sereda 1981). To assess the uncertainty in interval-velocity
derived by interpolating and contouring the 1D velocity functions. determination for the saturated zone, we calculated interval
The mean value of the interval velocity in the saturated zone is velocities for different time slopes Sx and stacking velocities

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
236 A. Becht, E. Appel and P. Dietrich

v2,STK varying within the limits of estimated parameter uncertain-


ties. We used the CMP at 12-m profile distance as an example for
these calculations. The model parameters v1, t1(0) and t2(0) were
fixed at values of 0.122 m/ns, 53 ns and 191.5 ns, respectively.
We estimated a parameter uncertainty for v2,STK of ±0.005 m/ns and
for Sx of ±0.5 ns/m, corresponding to an uncertainty in dip angle
of about ±1° for this example. The calculated interval velocities
are listed in Table 3. Computed interval velocities vary between
0.037 m/ns and 0.088 m/ns, where the actual determined value is
0.072 m/ns. In this case, the calculation of interval velocities is
thus very sensitive to the determined stacking velocity and the
time slope of the dipping reflector. Thus, small errors in the deter-
mination of stacking velocity and time slope may translate into FIGURE 13
large errors in interval velocity. The interval velocity in this exam- NMO correction and muting of some example CMP gathers: (a) CMP
ple is, therefore, rather inaccurately determined, since it is very gathers; (b) NMO correction; (c) muting.
sensitive to the accuracy of other parameters.
damage coherent reflection events. Thus, we seek a compromise
Stacked GPR sections between muting to reduce wavelet distortion and accepting large
The NMO correction removes the offset dependence of reflec- amounts of data to improve the signal-to-noise ratio. We defined a
tion traveltimes, so that each CMP trace can be treated as a zero- mute line constant for all CMPs by determining the zone in which
offset trace. Stacking of NMO-corrected CMP traces produces a distorted wavelets and non-horizontal arrivals are dominant in the
single zero-offset trace which shows an improved signal-to-noise NMO-corrected data (Fig. 13).
ratio. The maximum improvement of the signal-to-noise ratio Figure 14 compares stacked sections with different fold
that can be achieved is a factor of n1/2, where n is the number of orders: a constant-offset section (minimum offset), which would
stacked traces (Hatton et al. 1986). Thus, we may expect typically be obtained by a standard GPR survey, a 4-fold stacked
enhanced imaging of subsurface structures and an increased section using only traces with offsets of 1, 2, 3 and 4 m, and an
depth of investigation in multifold stacked sections. 18-fold stacked section (maximum fold) using all traces. The
A consequence of dipping reflectors and lateral velocity gradi- 4-fold stacked section represents a result that could be obtained
ents is that the true reflector points of a CMP do not coincide, so by simultaneous acquisition of four channels using an arrange-
that a smeared area of a reflector surface is sampled rather than a ment of one transmitter and four receiver antennae. This type of
common reflector point (assuming ray propagation). Therefore, data acquisition is nowadays possible with some commercial
reflector points are imaged incorrectly when NMO corrections GPR systems and is very time-efficient compared to the CRP or
are applied before stacking. More sophisticated processing CMP acquisition modes usually applied with single-channel
schemes, including dip-moveout correction or prestack migra- acquisition systems.
tion, allow for a more accurate imaging of reflector points (Russel The 4-fold stacked section shows significant improvements
1998; Yilmaz 1987). However, Greaves et al. (1996) pointed out compared to the constant-offset section. Structural details are
that, in general, the NMO correction is sufficiently robust to better imaged and the reflector continuity is substantially
improve data quality under most conditions. Since the gravel improved, especially for deeper reflectors. The depth of investi-
sheets in the area of investigation show comparatively small dip gation is increased by about 5–10%. Investigation depth and
angles of about 20–30°, we use the relatively simple NMO imaging of structures is further improved in the 18-fold stacked
approach for moveout correction of the CMP data. section. Reflectors that appeared indistinct in the constant-offset
Before stacking of NMO-corrected CMP traces, arrivals of section are clearly imaged and the depth of investigation is
direct waves must be eliminated or ‘muted’ to prevent artefacts. increased by more than 30%, compared to the constant-offset
Some additional points should be considered in choosing an appro- section. Evidently, it is necessary to stack all available traces for
priate muting window for optimum stack results. For a number of optimum results in the deeper part of the section.
reasons, it is also advisable to mute large-offset data of shallow
reflections. An undesirable effect of NMO correction is wavelet INTERPRETATION
distortion due to NMO stretching, which is most significant for For interpretation, the 18-fold stacked section in Fig. 14(c) was
shallow events at large offsets. As discussed earlier, NMO can devi- migrated with a phase-shift migration (Gazdag 1978) and travel-
ate strongly from hyperbolic in structurally complex areas with time was converted into depth. For these processing steps, we
lateral velocity gradients and when the magnitude of the offset used a 1D velocity model as shown in Fig. 15. The velocities of
approaches the reflector depth. In these cases, stacking of NMO- the unsaturated zone (0.122 m/ns) and the saturated zone
corrected CMP traces, based on flattening of hyperbolae will also (0.07 m/ns) correspond to average values determined from

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
Analysis of multi-offset GPR data 237

velocity analysis of the direct ground wave and cross-hole tom- rated to saturated zone between 40 ns and 70 ns. The parameters
ography, respectively. Very high gradients or steps in the veloc- of this transition zone were chosen, such that the reflector depths
ity model would lead to artefacts in the migrated depth-section; in the depth-section correspond to values obtained when using a
therefore, we used a linear transition of the velocity from unsatu- velocity model with a more realistic step-like velocity decrease

FIGURE 14
Stacked sections with different
fold orders: (a) Constant-offset
section with minimum offset of
1 m; (b) 4-fold stacked section
using traces with offsets 1, 2, 3
and 4 m; (c) 18-fold stacked sec-
tion using all traces.

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
238 A. Becht, E. Appel and P. Dietrich

FIGURE 15 from 0.122 m/ns to 0.07 m/ns at the groundwater level.


Velocity model used for migra- Obviously, the chosen 1D velocity model does not account for
tion and time-to-depth conver- lateral velocity gradients or different propagation velocities
sion. GW indicates the ground- within individual gravel beds. Therefore, this simplified velocity
water level. model introduces an error when converting traveltime to depth.
We estimate that this error is of the order of few decimetres,
based on test calculations with more complicated velocity
models derived from cross-hole tomography.
Figure 16(a) shows the migrated depth-section. We can iden-
tify four major continuous dipping reflectors: A, B, C and D. The
shallow reflector B is obviously distorted around the 8-m profile
distance, due to interference with the horizontal reflector related
to the groundwater level (GW). The deepest major reflector D is

FIGURE 16
Interpretation of the 18-fold stacked GPR
section. (a) Migrated depth-section. The let-
ters A to D denote major continuous reflec-
tors; GW is the groundwater level. (b)
Interpretation and correlation with drilling
core logs (gravel-type codes as in Table 1).
The width of the displayed core logs corre-
sponds to the dominant grain-size; lithologi-
cal units that can be correlated between
cores are shown in grey. Solid lines corre-
spond to major continuous reflectors, dashed
lines to minor reflectors.

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
Analysis of multi-offset GPR data 239

less pronounced in the right-hand part of the section due to signal poorly sorted gravels. Probably, this cobble-rich gravel bed is
attenuation. Other reflectors are less continuous and some show open-framework gravel which could not be identified in the
sinusoidal shapes. cores, due to material mixing in the process of core extraction.
Reflections of electromagnetic waves in low-loss geological In summary, major reflectors can be clearly correlated with
materials occur at interfaces between lithological units of con- lithological boundaries in the gravel sequence. Boundaries with
trasting dielectric permittivity. The dielectric permittivity of strong and sharp contrasts of porosity, such as transitions
clastic sediments is strongly related to water content. Thus, inter- between sand or sand-gravel and poorly sorted gravel and
faces of lithological units with different water content or, in the between open-framework and bimodal gravel, show a high
saturated zone, with different porosities act as reflectors. reflectivity. More gradual transitions, for example, between
According to a study by Huggenberger (1993), reflection coef- poorly sorted and open-framework gravel, are comparatively
ficients are largest between sand and open-framework gravel and indistinct.
between open-framework and bimodal gravels in unsaturated
conditions. In saturated conditions, the highest reflectivity DISCUSSION AND CONCLUSIONS
occurs at interfaces between sand and bimodal or poorly sorted We have processed, analysed and interpreted multi-offset GPR
gravels and between open-framework and bimodal or poorly data collected at a coarse-grained gravel aquifer within dipping
sorted gravels. Bimodal and poorly sorted gravels typically show foresets of a glacial delta. The maximum-fold stacked section
similar porosities and, thus, the dielectric permittivity is compa- shows enhancements regarding imaging of reflectors and depth of
rable in saturated conditions. In addition, the reflector strength is investigation, compared to a constant-offset section representing
dependent on the transition between lithological units. If the the result of a standard GPR survey. Our results confirm that the
transition is more gradual rather than sharp, reflections may be NMO correction is sufficiently robust to yield results with
relatively weak. For example, the groundwater level is often not improved data quality under marginal conditions when the antenna
imaged as a prominent reflector in GPR sections, because the offset is of the order of the reflector depth and the structure is
transition from the unsaturated to the saturated zone may be relatively complex with horizontal and dipping reflectors.
rather gradual in the presence of fine-grained matrix due to the Multi-offset data acquisition with a single-channel acquisi-
capillary fringe. tion system is very time-consuming and expensive. New com-
When correlating reflectors with core logs, we also must keep mercial GPR systems allow multichannel acquisition of, for
in mind that the accuracy of the core logs is limited. Material is example, four channels with different antenna offsets at the same
mixed during drilling and an accurate determination of litho- time. The time for data acquisition of such a 4-channel survey
facies types is clearly more difficult from core boxes than in would be equivalent to a standard constant-offset GPR survey.
outcrops. Therefore, there is some uncertainty related to the cor- We extracted a 4-channel data set with offsets of 1–4 m.
rect interpretation of lithofacies types, and the depths of the Reflectors are more clearly imaged and the depth of investiga-
boundaries are inaccurate to the order of several decimetres. tion is increased, when comparing the 4-fold stacked section
Figure 16(b) shows core logs of three drillings along the GPR with a constant-offset section. However, it was evident that all
section and an interpretation. The reflector A can be correlated traces must be stacked to obtain the best results in the lower part
with a layer of sand-gravel mixture intercalated into poorly of the section. We believe that the acquisition strategy of simul-
sorted gravels, showing up in drilling B3. Reflector B is related taneous acquisition of multichannel GPR data has great poten-
to the top of a layer of poorly sorted gravel, which is slightly tial. The optimum choice of offsets may depend on the data
coarser than the adjoining gravel beds. A very clear correlation is quality at the specific measurement site, the geological structure,
possible for the major reflector C. It represents the interface and the aims of the investigation, for example, whether the sur-
between highly porous open-framework and less porous bimodal vey should be optimized for maximum depth of investigation or
gravels. While this bottom boundary of the open-framework best resolution in intermediate depths.
gravel bed is well pronounced, the top is rather vague (dashed Velocity information about the subsurface is crucial for
line). This can be explained by outcrop observations: the bottom converting time-domain radargram data into depth-domain geo-
of open-framework gravels is typically a sharp lithological logical sections with correct reflector positioning. In addition,
boundary with underlying bimodal gravels (both units character- analysing propagation velocities in terms of hydraulic parameters
istically occur as couplets), whereas the top is often more gradu- would be of great benefit to hydrogeological studies and has been
al, due to a fining upwards within the open-framework gravel investigated by different authors (e.g. Greaves et al. 1996;
bed. Reflector D seems to be associated with the bottom of a Garambois et al. 2002). However, the determination of propagation
cobble-rich gravel bed in drillings B2 and B3 and with the top of velocities from surface CMP measurements is often a difficult task
a sand bed in B1. From flowmeter measurements, we know that and the accuracy of velocities based on NMO corrections is limited
this cobble-rich gravel bed is almost as permeable as open- (Tillard and Dubois 1995). In structurally complex areas with dip-
framework gravel and we may likewise expect a high porosity, ping reflectors and where velocity decreases with depth, which is
leading to a high reflectivity at the boundary with less porous, typical for GPR surveys of aquifers, the discrepancy between

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240
240 A. Becht, E. Appel and P. Dietrich

stacking and NMO velocity may become large. This was demon- REFERENCES
strated with a synthetic modelling study. In these cases, use of the Beyer W. 1964. Zur Bestimmung der Wasserdurchlässigkeit von Kiesen
und Sanden aus der Kornverteilung. Wasserwirtschaft-Wassertechnik
Dix equation or the more accurate 2D NMO approximation for
14, 165–169.
dipping layers results in unacceptable errors in computed interval Dix C.H. 1955. Seismic velocities from surface measurements.
velocities, because the small-offset approximation is not appropri- Geophysics 20, 68–86.
ate. To estimate propagation velocities, we considered the ray Fisher E., McMechan G.A. and Annan A.P. 1992. Acquisition and
geometry by forward-calculation of stacking velocities for the processing of wide-aperture ground-penetrating radar data. Geophysics
57, 495–504.
actual CMP geometry and iteratively determining the correct inter-
Garambois S., Sénechal P. and Perroud H. 2002. On the use of combined
val velocity. However, we showed that interval velocities are very geophysical methods to assess water content and water conductivity of
sensitive to the accuracy of determined stacking velocities and the near-surface formations. Journal of Hydrology 259, 32–48.
measured time slopes of reflectors. Errors in these parameters are Gazdag J. 1978. Wave equation migration with the phase-shift method.
greatly amplified. Therefore, we conclude that interval velocities Geophysics 43, 1342–1351.
Greaves R.J., Lesmes D.P., Lee J.M. and Toksöz M.N. 1996. Velocity
based on NMO-velocity analysis of CMP data are uncertainly
variations and water content estimated from multi-offset, ground-
determined parameters under such conditions. An interpretation in penetrating radar. Geophysics 61, 683–695.
terms of petrophysical parameters would be doubtful. Thus, we Gross R., Green A., Horstmeyer H., Holliger K. and Baldwin J. 2003. 3D
think that interval velocities based on surface CMP data are mean- georadar images of an active fault: efficient data acquisition, process-
ingful only under very favourable conditions. More reliable veloc- ing and interpretation strategies. Subsurface Sensing Technologies and
Applications 4, 19–40.
ity information and higher resolution could be achieved by meas-
Hajnal Z. and Sereda I.T. 1981. Maximum uncertainty of interval
urements with, for example, cross-hole radar tomography. velocity estimates. Geophysics 46, 1543–1547.
Major reflectors in the GPR depth-section were correlated with Hatton L., Worthington M.H. and Makin J. 1986. Seismic Data
boundaries between gravel types that show sharp contrasts of Processing: Theory and Practice. Blackwell Scientific Publications.
porosity, for example, the base of open-framework gravels or sand Heinz J., Kleineidam S., Teutsch G. and Aigner T. 2003. Heterogeneity
patterns of Quaternary glaciofluvial gravel bodies (SW-Germany):
beds within poorly sorted gravels. However, it was difficult to
application to hydrogeology. Sedimentary Geology 158, 1–23.
identify clearly the top of open-framework gravel beds, since the Hollender F., Tillard S. and Corin L. 1999. Multifold borehole radar acqui-
top is often a more gradual transition. Thus, the results show that sition and processing. Geophysical Prospecting 47, 1077–1090.
the geometry of gravel deposits can be investigated successfully Hubral P. 1976. Interval velocities from surface measurements in the
with GPR. It is, however, unlikely to map the thickness of indi- three-dimensional plane layer case. Geophysics 41, 233–242.
Huggenberger P. 1993. Radar facies: recognition of facies patterns and
vidual gravel beds, such as the hydraulically highly conductive
heterogeneities within Pleistocene Rhine gravels, NE Switzerland. In:
open-framework gravels in this field example. Braided Rivers (eds J.L. Best and C.S. Bristow), pp. 163–176.
In summary, multi-offset GPR data show significant improve- Geological Society Special Publication No. 75.
ments regarding reflector imaging and depth of investigation Kostic B. 2004. 3D sedimentary architecture of Quaternary gravel bod-
compared to standard constant-offset data. However, extracting ies (SW-Germany): implications for hydrogeology and raw materials
geology. PhD thesis, University of Tübingen, Germany.
reliable propagation velocities is difficult and may be achieved
Pipan M., Baradello L., Forte E., Prizzon A. and Finetti I. 1999. 2D and
only under very favourable conditions. A good compromise 3D processing and interpretation of multi-fold ground penetrating
between expenses in acquisition time and benefits in subsurface radar data: a case history from an archaeological site. Journal of
imaging is simultaneous data acquisition with a multichannel Applied Geophysics 41, 271–292.
GPR system. Thus, we believe that there is great potential for Russel B. 1998. A simple seismic imaging exercise. The Leading Edge
17, 885–889.
multi-offset GPR.
Shah P.M. 1973. Use of wavefront curvature to relate seismic data with
subsurface parameters. Geophysics 38, 812–825.
ACKNOWLEDGEMENTS Tillard S. and Dubois J.-C. 1995. Analysis of GPR data: wave propaga-
This work was supported by the German Science Foundation tion velocity determination. Journal of Applied Geophysics 33, 77–
(DFG) as part of the projects, Te 155/13-1 and Di 833/4-3. We 91.
Yilmaz Ö. 1987. Seismic Data Processing. Investigations in Geophysics
thank Dr. Jens Tronicke for help in data acquisition, Björn
No. 2. Society of Exploration Geophysicists, Tulsa.
Heincke for performing the migration, and Dr. Boris Kostic for
sharing his sedimentological expertise. We also thank the com-
pany Meichle & Mohr, Immenstaad, for access to the test site
and support in the gravel pit.

© 2006 European Association of Geoscientists & Engineers, Near Surface Geophysics, 2006, 4, 227-240

You might also like