Introduction to the processing of GPR-data within REFLEXW

In the following the processing of 2D-GPR data is discussed. Starting from the different
possibilities of data acquisition several filters are discussed later.

I. Data acquisition and import into Reflexw

I.1 import - general use

In order to use the acquired data within Reflexw

the data must be imported. This is done within the
2D-dataanalysis module using the option
file/import. Step by step procedure:
1. enter the module 2D-dataanalysis
2. enter the import menu using the option
File/Open/import. The REFLEXW-DataImport
menu appears (see figure at the right).
3. Make the following inputs:
input format: dependent on the acquistion system,
e.g. MALA-RD3
output format: new 16 bit integer or new 32 bit
floating point. The 32 bit floating point format
must be used is the original data have a higher
precision than 16 bit integer. This is often the case
for SEG2 or SEGY-data and for the newer GSSI-
data.
filename specification: original name for example
Choose X or Y as ProfileDirection and Y or X as
ProfileConstant.
Choose if the traceincrement and/or the
coordinates shall be read from the original data.
4. You may set the plot options which are also
stored with each imported file using the speed
option. Here you may choose between Point and
Wigglemode, ....
5. Activate the option Convert to Reflex. A fileopen
menu appears with the directory ASCII under your
project directory as the standard import path. You may
choose an original RAMAC file (RD3 or RAD-file)
from this import path or from any other directory. In any
case all necessary original files (for the MALA-RD3
format the RD3 and RAD) must be present. After having
choosen the wanted original file the data are converted
into the REFLEXW internal format and stored under the
path ROHDATA under your project directory. With the
option PrimaryFile activated the imported data are
automatically displayed into the primary window.

Sandmeier geophysical software - REFLEXW guide 1

I.2 different data acquisition modes

Normally the data will be acquired along 2D-profiles. The goal are equally spaced data with a start
and end coordinate and a traceincrement. This may be achieved by:

I.2.1 equally spaced data

When using a wheel or an equivalent method for an equally spaced data

acquisition single traces are only acquired at distinct spatial intervals.
These traces are combined into a 2D-line. The x-axis already represents
the distance. For the import within Reflexw enter file/import and
choose the necessary parameters. For Mala data the traceincrement has
been stored within the original data. Therefore the option read traceincr.
must be activated. The start coordinate in ProfileDirection (in this case
X) will not be changed. The entered end coordinate will be changed
according to the number of acquired traces and the original
traceincrement. If the option read traceincr. will be deactivated the
entered values for the start and endcoordinates will not be changed and
the traceincrement will be calculated from the coordinate values and the
number of traces. This is necessary if the original traceincrement has
not been stored (of course int his case the correct start/end coordinates
must be known).

Sandmeier geophysical software - REFLEXW guide 2

I.2.2 time based data

If the data are acquired based on a fixed time base, e.g. 100 traces every second, the correlation to
the distance is not given. In this case equally spaced markers (e.g. every 2 m) may be used for a
reinterpolation of the data. During the data acquisition these markers must be placed at distinct
intervals. A subsequent data processing step rescales the data from the time base to equally spaced
data.
Within Reflexw this is done using the
processing option mark interpol under
processing/TraceInterpolation/Resorting.
The option allows an interpolation of the
data in X-direction on the basis of
markers to be set manually or to be
automatically extracted from the data
(suboption ReadMarker). After having
activated the option a table
appears which allows the
interactive input of the marker
positions in traces (trace nr.)
and the distance between
successive markers (marker
inc.). The program
recalculates the number of
traces between two set
markers based on the wanted
trace increment and the current
set distance between the two
markers (marker inc.) using
the following formula:
calculated tracenumbers =
marker inc. / trace increment. The
marker increment must be predefined
within the filter parameter group and can
be individually changed for each marker
position within the table input. The
traceincrement must be a multiple of
each marker increment. The program determines the corresponding number of traces between all set
markers after the selection and executes a recalculation of the traces between successive markers. If
more traces within the original profile between successive markers are availabe, some traces are
omitted. If less traces are availabe, some trace are added within each marker part until the wanted
number of traces are reached (the traces are simply added and not interpolated between successive
original traces as such an interpolation may significantly change the shape of the onsets). The data
before the first marker and behind the last marker are ignored (exception if the option keep last
traces is activated), i.e. the program assumes that the beginning and the end of the profile are always
indicated by a marker. Please note that the data are modified by the interpolation. The trace with the
last marker position is also included. Therefore the total number of traces is increased by 1.

Sandmeier geophysical software - REFLEXW guide 3

I.2.3 GPS based data

If GPS-data are simultaneously acquired it is possible to

synchronize the GPR-data with the GPS-coordinates. The
original GPR data may either be time based or based on a
wheel. The different GPR-systems use different
synchronization types. Mala, Utsi, IDS or PulseEkko
generate a gps-file which contains both the tracenumber of
the GPR-file and the GPS-coordinates. For most data
acquisitions systems it is possible to automatically import
the gps-data into the Reflexw file during the import and to
perform a subsequent UTM-conversion. A
linear interpolation will be automatically done
where no GPS-data are present. The option
calculate distancies sums up the distance along
the gps-line and stores it into the Reflexw
traceheader. The GPS-coordinates may be
controlled and edited within the edit traceheader
tabella.

A subsequent processing step named make

equidist.traces under processing/
TraceInterpolation/ Resorting allows to
interpolate the non-equidistant data in such a
way that the resulting data are equidistant. The
non-equidistant data are resampled in x-
direction based on the filter parameter
trace incr. and the distance values stored
in the individual trace headers of each
trace. In addition the start distance and the
end distance
(starting and
ending position of
the new profile)
have to be
specified in the
given distance
dimension. By
default the start
distance and the
end distance are
determined from
the traceheaders.
By the manual
input you may
extract a distinct
part from the
profile.

Sandmeier geophysical software - REFLEXW guide 4

I.2.4 meandering data acquistion

Reflexw only accepts positive trace increment. If the profiles have been acquired in different
directions, e.g. using a meandering data acquisition, the
corresponding profiles must be flipped in distance direction.
This may be done directly during the import by entering a start
coordinate which is larger than the endcoordinate in profile
direction or using the processing option XflipProfile under
processing/traceinterpolation.
If GPS-data are present the flipping must be done after the
synchronization of the gps-data only as a processing step.

Sandmeier geophysical software - REFLEXW guide 5

I.3 display the data

After having done the import the data

are displayed using the standard plot
options. You may change these plot
options using the option Plot/Options.
Activating this option the Plot Options
menu appears (see figure on the right).

The main plot options for GPR-data

are:

- Plotmode
- PointmodeScale
- EnergyDecay
- Dewow
- AmplitudeScale

Plotmode: by default use Pointmode

for GPR-data.
PointmodeScale:
XYScaledPlot: the data are completely
plotted into the actual window
provided that the two scale options
XScale and YScale are set to 1. The
option may be used for small data (few
traces) or if you want to display the
complete dataset into the main menu (
the Zoom-options are available - see
also option Xscale and Yscale)). With no zooming and large datasets the display resolution is quite
poor.
PixelPerSample: the plotting size of each data point is given in screen pixels. Zooming or rescaling
is only possible in y-direction (option Yscale). The option might be useful for large data.
PixelsPerTrace: the distance between successive traces is given in screen pixels. The complete time
series of each trace is plotted corresponding to the size of the actual window. No zooming
possibilities are available. For example this option may be used for large data (many traces).
AmplitudeScale: this parameter controls the contrast of the data display.
Some filters may be applied during the display without going through a processing step:
EnergyDecay: most GPR-data are raw data without any gain in time-direction. Therfore activate this
option if no gain-filter (see chap. II) has been applied on the GPR-data in order to compensate the
energy decay with time.
Dewow: often GPR-data show a low-frequency content or DC-shift. With this option activated a
running mean value is calculated for each value of each trace which is subtracted from the central
point. Again use this option if no processing step like bandpassfiltering or dewow has been applied.

The following Figure show the data plotted using the options shown above.

Sandmeier geophysical software - REFLEXW guide 6

To be considered: Normally GPR data must be filtered (see chap. II). The main filters are:
- static correction to compensate for the time delay of the first arrival.
- y(time)-gain
- dewow or bandpassfiltering
- clutter reduction (e.g. background removal)

Sandmeier geophysical software - REFLEXW guide 7

II. Processing and filtering
The principal goal of data processing is to present an image which can be best possible interpreted.
The main purposes of data processing are:
- increase the signal to noise ratio (e.g. stacking, bandpass filtering, averaging)
- remove system induced irregularities (e.g. background removal, static correction)
- correct geometrical effects due to the data acquisition (e.g. migration)
Data processing can be classified as following:
- A-scan processing - the filter acts on each trace independently
- B-scan processing - the filter will be applied on the complete B-(2D-)scan and involves all traces
or a part of them. If a complete equidistant 3D dataset exists the processing may not be restricted to
one direction but may include data acquired within the complete xy-plane.
The mathematical background of the following standard data processing methods is quite
straightforward. The main problems consist in the best possible adaptation of the methods to the
type of data used and in the best possible use of the necessary parameters.

Sandmeier geophysical software - REFLEXW guide 8

II.1. Dewowing and standard bandpass filtering

Many GPR data show a significantly very low frequency component either due to inductive
phenomena or possible instrumentation restrictions. This low frequency range must be removed
before applying any other digital filter algorithms. There exist many different ways.
A simple dewow filter acts within the time domain. A running mean value is calculated for each
value of each trace. This running mean is subtracted from the central point. As filter parameter the
time range for the calculation of the running mean value must be entered which should be set to
about one or two principal periods. A possible static shift will also be removed using this filter.
Alternatives to the dewow filter may be a high pass bandpass filter working either within the
frequency or time domain or a simple subtract DC-shift filter if only a constant value shall be
removed.
The Figure shows an example of the dewowing
process. The left trace shows the original data,
the middle trace has been filtered using the
dewow filter with a filter length of 25 ns and
the right trace has been filtered using a
bandpass filter working within the frequency
domain with cutoff frequencies of 5 and 150
Mhz respectively and an adequate tapering
window.
The bandpass filter does not affect the signal
shape whereas the simple dewow filter slightly
changes the signal due to its non symmetrical
shape. Small precursors may occur (in some
cases also when using the bandpassfilter) which must be neglected when applying the next
processing step, the removal of the time base shift.
The bandpass filter may also be used in order to get rid of high frequency or monochromatic noise
(see Figure below).

Sandmeier geophysical software - REFLEXW guide 9

II.2 Time zero and time base shift correction

An exact definition of time zero is nearly impossible. It is not a constant value but depends on the
surface material type and the antenna set up configuration (see Figure below).

geometry sketch of a standard transmitter receiver configuration and possible raypaths.

.
The following Figure shows a comparison between the simple correction to the first onset (circles
within the Figure) and the dynamic correction (dots) under consideration of the source receiver
distance and the velocity of the medium. The reference builds the dynamic correction based on the
source receiver distance, the size of the air gap und the velocities of air and the medium therefore
taking into account the refraction of the omitted waves. If the velocity contrast is sufficiently high (>
2:1, left panel) the error using the simple static correction is small enough in order to ensure a
sufficient accuracy for the complete time range.

comparison of the errors for the time zero correction when using a static correction (open
circles) and a dynamic correction (filled rectangles) for a high velocity (left panel) and a low
velocity contrast (right panel)

Sandmeier geophysical software - REFLEXW guide 10

An automatic and stable static correction may be done either on the first negative, first zero crossing
or first positive peak. A correction to the first break position might be the best solution but may be
unstable in time due to variations of the electromagnetic properties of the underground in the near-
field of the antenna. An alternative may be the automatic correction to the first zero-crossing and
then performing a static shift to positive times with a signal length of the first peak.
In any case the picking of the reflections or diffractions and the subsequent depth calculation must
consider the chosen time zero method.
The time base of GPR measurements is also not exactly given and it may exhibit a significant drift
due to a temperature difference between the instrument electronics (especially concerning the
avalanche transistor effect) and the air temperature or damaged cables. Such a drift causes
misalignment of the reflections. The following Figure shows the application of the correct
max.phase filter under processing/satatic correction which automatically corrects to the first
positive onset within a distinct time window, makes a correction to the zero crossing and then
corrects the time zero to that zero crossing.

The subsequent filter move

starttime may be used in order
to remove all data in front of
the time zero.
Another cause of such a drift
may be the use of an air
coupled antenna with a
distinct distance from the
surface, e.g. a horn antenna.
This distance may vary during
the data acquisition. The
figure shows an example of
such a horn antenna
acquisition (top raw data, bottom time base corrected data). The filter correct max. phase with an
automatic correction of the time zero position has been used for this purpose.
A well-corrected time base is also very important for the interpretation of a 3D-dataset especially
when looking at the timeslices (C-scans) as different time bases may significantly destroy the
coherent character of the reflecting elements..

Sandmeier geophysical software - REFLEXW guide 11

II.3 Time varying gain

The waves will loose significant energy during travelling through the subsurface due to spherical
divergence and intrinsic and scattering attenuation. Therefore, these energy losses have to be
compensated. Several conditions have to be kept: The time series must have a zero mean value,
otherwise a significant DC offset especially at later times may occur. In addition the noise level at
greater times should be as small as possible.
It is possible to enter a manual gain value or to use a continuous gain function (see Figure below).
When manual gain values will be applied, rapid changes of the gain values should be avoided
because these may introduce unwanted artificial wavelets.
It is strongly recommended to use the same gain function for all profiles which shall be interpreted
together. This also holds true for 3D-data especially when looking at timeslices.

Sandmeier geophysical software - REFLEXW guide 12

II.4 Clutter reduction, background removal

GPR data are often contaminated by clutter. The clutter mainly consists of the GPR system noise,
ground bounce, soil roughness scattering and reflection signals from external anomalies. The clutter
mostly appears as nearly horizontal and periodic ringing.
Clutter reduction is therefore one of the most important challenges as especially deeper or weak
events are often completely masked by this clutter.
The GPR system based coherent noise ringing can be easily eliminated using a simple background
removal (subtraction of an average trace) filter as the statistical properties of the clutter have only
weak variations along the distance axis (see following Figure).

The situation is much more complicated if the statistical properties of the clutter vary along the
distance axis due to different ground coupling and/or due to subsurface scattering. In this case more
sophisticated methods must be used. Two-dimensional filters like fk-filter or Radon transform,
predictive and deterministic deconvolution or eigenimage processing techniques are the most used
ones. For all these methods the definition of the filter parameters must be adapted at the individual
situation in order to keep horizontal events but to guarantee a high performance of ringing
elimination. Especially deconvolution and eigenimage processing require careful setting of the
entered parameters. A good compromise between easy use and performance may often be given by
the fk-filter or the subtracting mean in combination with a notch filter if the noise exhibits
monochromatic characteristics.
The following Figure shows the comparison of the raw data which include both system induced
coherent clutter and incoherent ringing due to the subsurface conditions. The background removal
fails in the elimination of the incoherent clutter but both the fk-filter and the subtracting mean
within a moving trace interval seem to be a sufficient approach. In addition the background removal

Sandmeier geophysical software - REFLEXW guide 13

filter introduces some artefacts resulting from stronger ringing within some ranges, e.g. between 160
m and 200 m, which has been smoothed over the complete distance range.

comparison of different methods for incoherent clutter reduction. The upper left panel shows the
original data, the upper right panel the background removal filtered, the lower left panel the
application of the fk-filter and the lower right panel the use of the subtracting average filter.

Sandmeier geophysical software - REFLEXW guide 14

II.5 Migration

The waves coming to the receiver will be acquired vertically along the acquisition line and therefore
do not represent the correct positions of small scale diffractors or sloped reflectors. The goal of the
migration is the downward continuation of the acquired wavefield to their origin. The base for this
wavefield continuation is a given depth velocity model. During the migration process diffractions
will be concentrated and dipped layers will be moved to their correct place. The precondition for the
migration is a good knowledge of the underground velocity field. The velocities may be easily
examined using the interactive velocity adapation within Reflexw.

Standard migration is done for 2D-data. Fast algorithms exist for a constant velocity field (e.g. fk-
migration - see Figure above) but also the more sophisticated methods like the Finite Difference
approximation of the one way equation can be applied on standard PC’s with reasonable computer
time consumption. These methods also allow a 2D-velocity distribution (see also chap. II.7 - time-
depth conversion). The following Figure shows a synthetic example (FDTD method) including
diffractions and steep reflections. The model shown within the left panel serves as the base for a
forward simulation of a ZO-section (middle panel). The small elements within the first layer
produce diffractions which interfere each other that does not allow the individual identification. The
trough at about 10 m is characterized within the ZO section by two diffractions and a reflection
from the bottom. It is evident that only after migration (right panel) the structure elements have been

ZO simulation (middle panel) for a model including small diffractors and steep reflectors (left
panel). The right panel shows the migrated ZO section.

Sandmeier geophysical software - REFLEXW guide 15

shifted to their real location and the diffractions have been concentrated so that they can be
distinguished from each other.

Migration is most useful if timeslices (C-

Scans) will be produced for the subsequent
interpretation. The Figure on the right shows
both the raw data (left panel) and the
migrated data (right panel). Again the strong
energy concentration within the migrated
data is evident which leads to a much better
discriminability of the individual elements.

A 3D-migration is useful for a coarse 3D-datagrid with equal spacing in x- and y-direction. Due to
the large computer time for a complete 3D-migration methods with a constant migration velocity
(e.g. Kirchhoff migration) will be used by default.

Sandmeier geophysical software - REFLEXW guide 16

II.6 topographic correction

The GPR-profile normally exhibits the same start time for all traces. Therefore it does not reproduce
the surface topography. As a consequence reflections which are more or less flat in reality may
exhibit significant curvatures due to the surface topography and the concomitant varying layer
thickness above.
If the topographic values have been
stored within the traceheaders (e.g.
when a GPS-system has been used) the
plotoption correct header elevations
may be used in order to display the
data including the topography (see
Figure right).
There are different other possibilities
to perform a topographic correction
within Reflexw. Al these methods may
be found under processing/static
correction/muting and they act as a
filter (the data will be changed - filled
up with zeroes above the topography).
The easiest ways is the option static
correction with the suboption move to
positive times. The topographic values
may be entered interactively by mouse
or within the table (two-way traveltime
values).
Another possibility is the option
correct 3D-topography. The elevation
or depth values may have been stored
within the Reflexw traceheaders (see above) or may be read from an ASCII-file. Different
coordinate systems may be used for storing the topographic values.

Sandmeier geophysical software - REFLEXW guide 17

II.7 time depth conversion

The GPR-profile respresents a two-way traveltime section. All

processing steps are normally based on this scale (even the
correct topography option). In order to convert the time axis to
a depth axis a velocity distribution is needed.
Reflexw allows to view a depth or elevation axis at the right
hand side in addition (plotoption DepthAxis activated) based
on a constant velocity or based on 1D-velocity distribution
(option non linear depth axis activated - only variations of the
velocity with depth allowed).
A more sophisticated possitility is given as a filter processing step timedepth conversion under
processing/migration/time-depth conversion. There are several possibilities to get information about
the velocities:
- from the curvature of existing diffractions within normal zero offset profiles
- from CMP measurements if a layered medium is given
- from borehole measurement and subsequent comparison of these information within the zero
offset data
- from the literature if the material of the underground is known
The easiest way is the use of a constant velocity. This does not affect the shape of the signals within
the radargram. If using a time varying or even
time-distance varying velocity distribution for
the time-depth conversion the signal shape
may be changed quite strongly. This must be
considered for the subsequent interpretation.
For most cases, especailly if a layered model
shall be the base for the time depth conversion
the CMP-analysis might be the right choice
even if no CMP-meausrements are present.
On the right side a 1D-velocity model has
been created based on different informations
like diffractions and cores. This velocity
model may be used in order to generate a
time-depth converted section (see Figure
below).

Sandmeier geophysical software - REFLEXW guide 18

II.8 profile energy balancing

A great problem is the often non-uniform energy feeding-in along a 2D-profile or for different 2D-
profiles which shall be interpreted together (e.g. creating time slices). The causes may be varying
ground conditions, equipment changes, use of multi antenna systems or differences in the field
acquisition. Whereas variations along the profile caused mainly by different coupling conditions can
be quite easily compensated using a trace normalization or a gain function in profile direction, more
sophisticated methods must be used when dealing with 3D-data especially when the interpretation is
based on time slices.

The Figures above show an example of a multi-grid survey using a 2-antenna system with different
energy characteristics. The left panel shows 2 parallel 2D-raw profiles (0.05 m increment) indicating
the different energy content. The right panel shows two time slices. The left time slice is based on
the raw data, for the right time slice an energy compensation has been applied at first. The
horizontal stripes on the left time slice are due to the different energy characteristics of the used 2
antennas. They may lead
to misinterpretation.
The right Figure was
acquired with only one
antenna but at different
conditions. Therefore
some areas show
different energy
distributions (left panel).
A careful normalization
allows an uniform
picture (right panel).

Depending on the used equipment and the existing ground conditions often not only a uniform
factor but a time varying curve must be determined for the compensation of the different energy
feeding-in. This may lead to an amplitude decrease of reflections which are only present in some
profiles (e.g. pipes which are orientated parallel to the acquired profiles). Therefore, such a
compensation must be used very accurately.

Sandmeier geophysical software - REFLEXW guide 19

II.9 suppress multiples

Sometimes surface multiples may occur if a near surface reflector is present with a strong velocity
contrast. Possible deeper reflection may be hidden by these multiples. The option suppress
multiples under processing/static correction allows to suppress those multiple reflections from the
surface. For that purpose the first reflection must be picked (preferably shifted to the very first
onset). The picks must be saved under the same filename as the current profile filename.
If using the processing option suppress multiples multiple reflection are first flattened based on the
picked primary reflection and then the subtracting average filter is applied within a choosable time
range around the flattened multiple.

Sandmeier geophysical software - REFLEXW guide 20

II.10 F-K filter

Unwanted reflections from the borders of the investigation medium are often characterized by a
distinct slope which corresponds to medium or air velocity. Those structures can be easily removed
using a multichannel filter.
The most popular filter is the
so-called frequency-
wavenumber (F-K) filter
which works within the
frequency wavenumber
range.
The Figure on the right
shows the raw data (left
panel) including a distinct
side reflection and the
filtered data (right panel). It
must be kept in mind that
other elements showing the
same slope will also be
removed using such a filter.
In most cases tapering must
be used in order to avoid
artefacts.
Another application of the fk-filter is the removal of horizontal stripes, e.g. due to a ringing (see
also chap. II.4). The definition of a velocity range to be removed or kept allows the elimation of
structures characterized by these distinct slopes (corresponding to velocities within time-distance
range). The horizontal stripes for example are located arount small kx values within the kx-range
corresponding to very high velocities. The Figure below shows the raw data (top panel) and the fk-
filtered data (lower panel) based on the filter parameters defined within the fk-filter menu. The
resulting filtered fk-range is displayed on the right. The clutter is quite well reduced.

Sandmeier geophysical software - REFLEXW guide 21

II.11 deconvolution

A major problem in signal interpretation is the lack of resolution of overlapping events due to the
reverberation character of the signal. The main purpose of the deconvolution is to invert the
convolution process of the medium impulse response and the outgoing signal. The ideal outcome of
the filter is again the medium impulse response. Although this ideal cannot be usually achieved,
many different methods have been developed for different preconditions. One filter is named
Wiener filter which minimizes the differences between output and desired result. Other methods
work as direct inverse filter. In general deconvolution techniques are not very well suited for GPR
applications as the main preconditions like minimal phase, lag time zeros are normally not satisfied.
In many cases predictive deconvolution techniques or wavelet shaping may lead to better result. A
suitable filter strongly depends on the characteristics of the signal.
The following Figure shows an application of a wave shaping filter. The wave shaping filter allows
to convert the characteristic waveform of the profile to a new desired one. It is obvious that after the
shaping filter all signals exhibit a much clearer and sharper form leading to a better resolution in
time direction. The precondition is that a characteristic waveform ca be found which fits over the
complete time-distance range. Due to different coupling conditions and waveform changes during
the propagation this precondition is quite rarely satisfied. The spiking filter is a special case of this
filter which is designed to compress as much as possible the original wavelet into a spike (uniform
frequency distribution).

application of a wavelet shaping filter (left panel raw data, besides the filtered data, right panel:
one trace of the two datasets respectively).

The main goal of the predictive deconvolution is the suppression of multiples. The desired output is
a time advanced version of the input signal. To suppress multiples one has to choose a lag
corresponding to the two-way-traveltime of the multiple.
In the following synthetic radargrams have been created using a forward Finite Difference Time
Domain method (FDTD method) in order to examine the effectiveness of a predictive
deconvolution method for multiple and ghost removal.
The following Figure (left panel) shows an example with strong multiples coming from a very near
surface interface with very high velocity contrast. The used predictive deconvolution yields quite
good results (middle panel). The reverberating character of the signal could be reduced to a sharp
signal with only two maxima both for the primary onset and the reflections and therefore the
distinguishability has been significantly improved.

Sandmeier geophysical software - REFLEXW guide 22

 application of the predictive deconvolution (middle panel) on a synthetic radargram
with strong multiples (left panel). The right panel shows one trace of these datasets
respectively.

Some further investigations had been done introducing a ghost by bringing in a strong reflector
above the receiver line. The amplitudes of the ghost had been varied. Whereas the weak ghost can
be quite well eliminated using the predictive deconvolution, no good results have been achieved for
a ghost with amplitudes similar to those of the primary onset (see Figure below). After having
applied the predictive deconvolution the ghost is still visible although with smaller amplitudes
(middle panel). In addition the deconvolution process produces multiples.

application of the predictive deconvolution (middle panel) on a synthetic radargram

with strong ghosts (left panel). The right panel shows one trace of these datasets
respectively
In summary it can be said, therefore, that the predictive deconvolution method used for this
investigation was able to reduce multiples and weak ghosts. The signals presented within the
synthetitic seismograms are minimal phase. The effectiveness of the deconvolution will be lower if
this does not hold true or if reverberations are present in addition. Therefore, in reality the results of
the predivitive deconvolution may be significantly poorer.

In any case the methods always need an intensive adaptation of the filter parameters and a manual
check for possible artefacts.

Sandmeier geophysical software - REFLEXW guide 23