Вы находитесь на странице: 1из 4

AVO analysis of shallow seismic horizons: effect of accuracy and uniformity of the effective

source wavelet
Ranajit Ghose*
Delft University of Technology, The Netherlands
Summary
The amplitude of the shallow reflection events, and in
particular, the variation of reflection amplitude as a
function of offset (AVO) can provide useful information
about the lateral variation in the shallow subsurface. In this
research we have investigated the effect of accuracy and
shot to shot uniformity of the source wavelet on the result
of AVO analysis. For small vibrators, it is possible to
monitor accurately the effective source signature, including
the effect of source-ground coupling. This allows us to
deconvolve the raw vibrograms using the individual source
signature. For each shot the source signature can be
estimated separately and hence the effect of shot to shot
variation can be effectively removed by deconvolution. We
compare our result with that when the elimination of source
signature is not accurate and consistent (e.g., in case of
cross-correlation of vibrograms). We observe remarkable
improvement in the result of AVO characterization of soil
boundaries, with correct source signature deconvolution.

amplitude variation of the reflected signal that is caused by


variation of the effective source signature. On shallow
shear wave data, we show the improvement in the result of
AVO characterization of the subsoil boundaries, when the
variation of effective source signature is eliminated.
Source signature elimination: possibility for vibrators
Small, electromagnetic vibrators have been developed
lately for controlled generation of high-frequency P and S
waves for shallow, engineering applications. Because of
small energy, for such vibrators it is possible to monitor the
source motion and hence the generated (effective)
groundforce (including the effect of source-ground
coupling or the radiation impedance) quite accurately for
the entire frequency bandwidth of the signal. Because a
good groundforce estimation is possible, we can compress
the raw vibrograms by deconvolution. On field data, we
have recently observed that compression by deconvolution
offers better separation of the shallow reflection events than
compression by cross-correlation (Ghose, 2002).

Introduction

Variation of the effective source signature is more critical


for shallow, high-resolution seismic data than for deeper
data because of (1) the relatively high frequency content of
the data, (2) closely spaced reflectors in the shallow
subsoil, and (3) greater influence of the near-surface effects
on the signal. In the present research we investigate the

We have estimated the source wavelet for 4 different


approaches of vibrogram compression, viz. (a) crosscorrelation using the sweep signal that is amplified and fed
into the shaker, (b) cross-correlation using the groundforce
(Fg) estimated from the measured reaction-mass and
baseplate accelerations (Fg = mrar+mbab, where r and b
indicate respectively reaction-mass and baseplate, and m
and a are their respective mass and acceleration), (c)
dephasing using Fg, and (d) deconvolution using Fg. This
is illustrated in Fig.1. After cross-correlation, the source
Power
Spectrum

Source Wavelet: Shot # 10


100

120
20

sweep
correlation

groundforce
deconvolution

100
Time (msec)

120

150 300

150 300
(Hz)

15
0

0
20

15
0
15
0

0
20

0
80

150 300

0
20

groundforce
correlation
groundforce
dephasing

Phase
Spectrum

(deg x -1000)

80

(dB)

Progress in high-resolution seismic reflection method now


allows us to map the shallow soil structure in fair details.
Particularly the use of shear waves in soft soil has proven to
be quite efficient in order to achieve high resolution,
because of the low velocity of the shear waves in watersaturated, soft soils. 2-way time of the seismic reflections
offers information of the structure and/or the seismic
velocity and their lateral variation. More recently, the
variation of reflection amplitude as a function of offset
(AVO) or incidence angle has been used to infer the nature
of the subsoil boundary and local material property (Ghose
and Goudswaard, 2000), which are of prime concern to the
engineers. Observed variation of reflection amplitude may
be caused by geological reasons, but may also be induced
by variation of the effective source signature from shot to
shot or by variation of receiver coupling from geophone to
geophone. There has been little progress so far to
deterministically correct for the amplitude variation due to
variation in source or receiver response.

150 300
(Hz)

15

Figure 1: Source wavelet, and its (relative) power and phase


spectra for 4 different approaches of vibrogram compression.

Source wavelet and AVO analysis

wavelet still contains the information of the sweep and the


vibrator response including vibrator-earth coupling,
harmonic distortions and the eigenfrequencies. When the
sweep is used for cross-correlation, the power spectrum is
flat in the frequency range of the sweep but the power is
proportional to the actual power of the groundforce. After
dephasing using the groundforce, the phase information of
the source is removed but the amplitude information still
remains. After deconvolution, both amplitude and phase of
the source are removed, and hence the power spectrum of
the source wavelet after compression becomes flat within
the frequency range of the sweep (where signal is larger
than noise), the power represents the true source power.
Field experiment
The issue of source coupling is more problematic for S
wave than for P wave. We have conducted a shallow Swave reflection experiment using the horizontal vibrator
source. 48 single horizontal geophones (28Hz) were used.
Geophone interval was 0.5 m. Source interval was 1.0 m.
To record the SH wave, the horizontal geophones were
oriented in crossline direction. For the entire experiment,
the receiver layout was kept fixed, and only the source was
moved. The frequency bandwidth of the vibrator sweep was
50-300 Hz, the sweep was linear with length 3.5 sec, the

Fig.2 illustrates four representative raw shot gathers and the


same shot gathers after noise filtering and muting. For this
illustration, groundforce deconvolution has been used to
compress the vibrograms. Preprocessing involved trace
editing, muting, geometrical spreading correction, f-k
filtering, and bandpass filtering. Eliminating the surface
waves was the main challenge. Very clear, shallow S-wave
reflection events can be seen in the shot gathers. It is also
apparent from the appearance of the reflection events in the
preprocessed shot gathers that there is lateral variation.
The source wavelet was estimated at each source location
in the field, using various approaches for compression of
raw vibrograms. Fig.3 shows in time domain the lateral
variation (at 3 m interval) in amplitude of the source
wavelet for groundforce cross-correlation (imperfect
elimination of the source signature) and groundforce
deconvolution (almost perfect elimination of the source
signature). It is clear that the lateral variation of the source
wavelet is conspicuous if the effective source signature
including the effect of source coupling is not correctly
removed (Fig.3(a)). The amplitude of the source wavelet
becomes laterally quite uniform when the effective source
signature is correctly removed by deconvolution of the

S-wave shot gathers: raw


0

FFID 55

FFID 71

FFID 75

FFID 102

100

100

200

200

S-wave shot gathers: preprocessed


0

FFID 55

FFID 71

FFID 75

FFID 102

100

100

200

200

Figure 2: Shallow S-wave reflection experiment: representative shot


gathers (a) before and (b) after preprocessing.

Time (msec)

(b)

Time (msec)

Time (msec)

(a)

Time (msec)

record length was 4.0 sec. Data sampling was at 1 kHz.

Source wavelet and AVO analysis

(implying that the lateral variation of the source signature


has been sufficiently taken care of). Figs. 4(a2) and 4(b2)
show the 48-channel average amplitude spectrum of the
raw geophone data; the difference between these two
figures is entirely due to the difference in the source
wavelet present in the respective shot gathers. Strikingly,
the variation in the source wavelet is large enough to be
clearly discriminated by the geophones. The surface
condition at this experiment site is quite flat and uniform,
nevertheless the shot to shot variation in source coupling is
large for applications involving true reflection amplitude.
Laterally consistent, correct elimination of the source
signature reduces the lateral variation in the average
geophone spectrum (compare Fig.4(a2) with Fig.4(b2)).

(a) wavelet: Fg cross-correlation


source location (m)

70

12

15

18

100
(msec)
130

(b) wavelet: Fg deconvolution


source location (m)

70

12

15

18

100
(msec)

AVO analysis of shallow seismic horizons


130

groundforce (Fig.3(b)). This is crucially important in


applications involving amplitude variation of seismic
reflections, particularly for shallow seismic datasets
characterized by a relatively higher frequency content and
closely spaced reflection events.
Fig.4 relates the detected lateral variation in the effective
source wavelet to the variation in the geophone data. Figs.
4(a1) and 4(b1) illustrate the amplitude spectrum of the
source wavelets shown in Figs. 3(a) and 3(b) respectively,
i.e., after groundforce cross-correlation and deconvolution.
After cross-correlation, the wavelet spectrum is quite
different from shot to shot, particularly at low frequencies
(50100 Hz, as indicated by the box) where the source
coupling effect is particularly problematic. After
deconvolution, the wavelet spectrum is flat between 50 and
300 Hz (the sweep range) and is laterally uniform
(a1) correlation: wavelet

(a2) correlation: 48-ch. av. geophone

source location (m)


6

12

source location (m)

15 18

12

15

We have tested the effect of lateral variation of the


effective source wavelet on the AVO analysis of shallow
seismic horizons. Fig.5(a) shows the geology at our test site
derived from soil sampling and CPT. Average S-wave
velocity (Vs) in such sand and clay layers is known. An
approximate Vs model was assumed using the independent
information of geology. A finite difference (FD) synthetic
shot gather, with the same source-receiver geometry as our
field shot gather, was calculated (Fig.5(b)). The purpose of
this modeling was to check if the events seen in the field
data can be related to independently known geology. Since
we were interested in arrival time here, an acoustic FD
approach (Kelly and Marfurt, 1990) was sufficient. Fig.5(c)
shows the corresponding field shot gather after muting and
CPT data

(a)
.0

Rf (%)

fs (MPa)

.1

.2

.3

.4

12

16

10 8 6 4 2 0
20

24

soil
layer
boundaries

interval
Vs
model
(m/sec)

28

qc (MPa)

135 155 175


sand-fill

Depth (m)

Figure 3: Lateral variation of source wavelet


for Fg correlation and Fg deconvolution.

fs
4

A
qc

peat and
clay

Rf

B
sand

18
8

100

100

200

200

300 (Hz)

300 (Hz)

peat and
clay

10

sand

FD synthetic
shot gather

(b)

C
D

(c) field data: S-wave shot gather


source at CPT location

100

100

12

15

18

12

15

18

100

100

200

200

300 (Hz)

300 (Hz)

Figure 4: Lateral variation of the source wavelet spectrum related


to the variation of the 48-channel average geophone spectrum.

Time (msec)

source location (m)

source location (m)


1

Time (msec)

(b1) deconvolution: wavelet (b2) deconvolution: 48-ch. av. geophone

200

200

300

300

Figure 5: (a) CPT data, layer boundaries from soil testing and CPT, and
the interval Vs model, (b) synthetic shot gather, (c) field shot gather

Source wavelet and AVO analysis

Time (msec)

42

52

CMP #
62

72

32

82

40

80

120

82

80
5m

Horizon B
0

Intercept

+0.6

Horizon B
0

Horizon A

Horizon A

(a)

72

120

5m

-0.6
-3.0

CMP #
52
62

40

+0.6

Intercept

42

0
Time (msec)

32

0
Gradient

+3.0

(b)

-0.6
-3.0

0
Gradient

+3.0

Figure 6: Brute stack and AVO intercept-gradient crossplot for two seismic horizons (A,
B) for source signature elimination by (a) Fg cross-correlation, and (b) Fg deconvolution.

filtering. The strong reflection with 2-way zero-offset time


(t0) around 35 msec (consider the trough) is from the sandclay boundary located around 2.2 m. The reflection with t0
around 55 msec (consider the crest) is possibly the sandclay boundary at around 5.2 m; we see the same event in
the synthetic data. The AVO analysis was carried out for
these two very shallow horizons.

deconvolution can significantly reduce the source to source


amplitude variation, and hence homogenize the effective
source wavelet. The improvement in the result of AVO
analysis can be remarkable. More research on geological
interpretation of the variation of seismic reflection
amplitude in the shallow subsurface is required to highlight
the practical implication of our finding.

True amplitude (brute) stacks were created using 1-D


stacking velocity (corresponding to the interval velocity
shown in Fig.5). Two horizons (A, B) were chosen, as
shown in the upper stack sections in Fig.6. For each CMP,
AVO analysis was carried out at these two horizons. Our
objective was to evaluate the difference in AVO analysis
results when the source signature is removed by two
different ways. The gradient-intercept plots are shown, for
source signature elimination via cross-correlation of Fg and
via deconvolution of Fg. The raw data, preprocessing
parameters, the stacking velocity field, and the chosen
horizon for AVO analysis are identical between Fig.6(a)
and Fig.6(b), the only difference is in the source wavelet. It
is clear that when the source signature is removed via
deconvolution the clustering in the intercept-gradient plot
for a given horizon is significantly tighter than when the
source signature is removed via cross-correlation.

References
Ghose, R. and Goudswaard, J.C.M., 2000, Relating
shallow, S-wave seismic to Cone Penetration Testing
(CPT) in soft soil: a multi-angle, multi-scale approach, 70th
Ann. Internat. Mtg., Soc. Expl. Geophys., Exp. Abstr.
Ghose, R., 2002, High-frequency shear wave reflections
from shallow subsoil layers using a vibrator source: sweep
cross-correlation versus deconvolution with groundforce
derivative, 72nd Ann. Internat. Mtg., Soc. Expl. Geophys.,
Exp. Abstr.
Kelly, K.R. and Marfurt, K.J., 1990, Numerical modeling
of seismic wave propagation, Geophysics Reprint Series,
No. 13, Soc. Expl. Geophys.
Acknowledgments

Conclusions
Our research shows that if a good estimate of the source
signature (including source coupling effects, harmonic
distortions, and eigenfrequencies) can be made for each
shot at every source location individually then deterministic

We acknowledge the support of the Dutch Technology


Foundation (STW Grant No. DAR.5761). We thank
Gemeentewerken Rotterdam to allow us to do the field
experiment at their site.

Вам также может понравиться