Phase Decomposition As A DHI in Bright Spot Regimes A Gulf of Mexico Case Study

Phase Decomposition as a DHI in Bright Spot Regimes: A Gulf of Mexico Case Study
Umberto Barbato*, Lumina Geophysical , Oleg Portniaguine, Lumina Technologies, Ben Winkelman, Talos
Downloaded 11/01/17 to 132.236.27.111. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Energy, John Castagna, University of Houston
Summary the amplitude/phase/frequency relationship. Following

Meza’s approach, the attribute is applied on a 3D seismic
The study of phase dependent amplitude volumes can be an dataset from the Gulf of Mexico, where the target plays are
important tool in the interpretation of amplitude anomalies, mainly bright spots for light oil-prone sands. A zero phased
as hydrocarbons can change the odd component of the far stack (25° to 35°) is used as the input data. In this case,
seismic response of thin layers. The original spatial one pay and one wet well, on the same target sands, with
distribution and lateral extent of amplitude anomalies may similar thicknesses will be reviewed. This presents a
vary after phase decomposition, allowing the interpreter to scenario where the reservoir properties are similar, and the
better assess and prioritize target amplitudes. The thin layer main difference being the presence or absence of
requirement can be achieved by high-cut filtering the data hydrocarbons.
(increasing the tuning thickness) or running the attribute in
selected frequency-dependent amplitude volumes from Synthetic Modeling of Frequency, Phase, and
spectral decomposition. The definition of the necessary Hydrocarbon Effects
frequency content for the seismic data can be obtained by
building synthetic frequency gathers from well data and Phase decomposition does not use well information to
applying phase decomposition on these. A Gulf of Mexico produce phase components; however, synthetic traces
case study shows how phase decomposition changes generated from well logs can provide the required
amplitude maps and can be effectively used as a direct knowledge to establish the amplitude/phase/frequency
hydrocarbon indicator. relationships considered desirable. Applying spectral
decomposition to a well’s synthetic trace produces a
Introduction frequency gather that decomposes the trace into its
frequency dependent behavior. On the other hand, applying
Phase decomposition is a seismic attribute introduced by phase decomposition to this synthetic gather produces a set
Castagna et al., (2016) which allows us to separate the of frequency dependent phase component gathers. For
amplitude response of the seismic trace into its simplicity we will show here only the even and odd
corresponding phase components. An analogy can be components. These are a combination of the zero and 180
drawn to spectral decomposition, in which different events degree event phases (for even) and -90 and 90 degree
have different tuning frequencies; a similar approach can be phases (for odd). Given that we are studying amplitude
taken with phase decomposition, where amplitudes are anomalies in low impedance hydrocarbon bearing sands we
mapped as a function of phase. Phase decomposition can expect the hydrocarbon effect to be manifested in the odd
be used as a DHI. As Meza et al. (2016) explain “This is component.
based on the idea that, for thin layers, the “hydrocarbon
effect” or change in amplitude caused by the addition of The far-trace synthetic results for the area of interest are
hydrocarbons to a brine-filled thin layer, is -90 degrees shown in Figures 1 and 2 where we observe the frequency
phase-rotated with respect to the wavelet”. The output of gathers for the synthetic seismograms and their even and
phase decomposition includes amplitudes at specific odd components in two wells. The gathers are generated by
response phases, called phase components. In the case of applying the Continuous Wavelet Transform (CWT) (e.g.,
low-impedance thin gas or light-oil sands (bright spots) Chakraborty and Okaya, 1995), followed by phase
amplitude anomalies corresponding to gas saturations are decomposition. The frequency gather is labeled as ‘input’
expected to appear in the -90 degrees component. It is in both images, while the even and odd components from
important to remember that the seismic trace is amplitude, phase decomposition are labeled as such (‘even’ and
phase, and frequency dependent. The thin layer effect is ‘odd’). We can readily observe how amplitudes change
observed when the layer is below tuning, therefore, the with frequency and event phase. Two target sands are
frequency content of the seismic data must be monitored labeled (1 and 2). As the water saturation log shows, these
for targets with variable thicknesses. Thankfully, seismic are pay sands in Well-1 (black arrows) and wet in Well-2
layers can be made artificially thin by either low pass (purple arrows), and are both low impedance sands. The
filtering the data or using spectral decomposition frequency tops and bases of the sands are highlighted by the dashed
dependent amplitude volumes as the input for phase lines, and their thicknesses vary from 15 to 20 ft for Sand-1
decomposition, allowing the interpreter to better establish (from Well-1 to Well-2) and from 80 to 100 ft for Sand-2.
© 2017 SEG Page 3976

SEG International Exposition and 87th Annual Meeting
Phase Decomposition as a DHI in Bright Spot Regimes
Reviewing these figures, we can conclude that both sands Application in the 3D Seismic Far Stack
are below tuning at 20 Hz, where we observe the energy
moving towards the odd component in both cases, The original seismic has a dominant frequency between 20
rendering the sands below seismic resolution. and 30 Hz. As at this frequency content the thick sands are
resolved, a high cut bandpass filter (20,40 Hz) is applied
The acoustic reservoir response of the brine saturated sand seismically thinning these layers, so as to better reveal the
is an impedance drop for both cases, which is observed in phase-related anomalies. Both datasets become the input
the odd component. The pay well shows a clear amplitude for phase decomposition, and their even and odd responses
anomaly in the input frequency gather, which, after phase are compared.
decomposition, becomes evident in the odd component
(Figure 1). The synthetic trace of figure 2 shows relatively
bright amplitude responses in the input as well but the
energy is split between even and odd components after
phase decomposition. Given that the synthetics are built
with the same wavelet, the direct amplitude comparison
provides a potential signature of a DHI.
Figure 1. Synthetic frequency gather and its phase decomposition response for Well-1. Black arrows show the seismic dominant frequency
required to place the target sands below tuning for the pay well
Figure 2. Synthetic frequency gather and its phase decomposition response for Well-2. Purple arrows show the seismic dominant frequency
required to place the target sands below tuning for the wet well
© 2017 SEG Page 3977

Figure 3 shows an amplitude extraction along a horizon potentially wet. However, the target sands are not below
interpretation of Sand-2. An arbitrary line is shown (in red) resolution. For the high cut filtered data we can observe
which corresponds to the vertical display sections on Figure that the amplitude anomaly on Well-2 moves towards the
4. The map shows 2 clear amplitude anomalies that were even component (Figure 5e) and the bright spot that Well-1
drilled (wells 1 and 2). crosses moves towards the odd component (Figure 5f). It is
evident that separating the amplitude response in its phase

components can introduce a new interpretation point of
view. Figure 4f shows that sand-2 continues to have
potential for hydrocarbon presence. While the position of
Well-2 is structurally favorable, the sand appears to be
anomalously bright in the odd component downdip from
the well location, suggesting the possible presence of
hydrocarbons downdip which would require stratigraphic
discontinuity. Note that the seismic thinning via bandpass
filter was critical in order to achieve the expected
separation of phase dependent amplitude, highlighting the
necessity of awareness of target thicknesses. Understanding
the relationship between amplitude, phase, and frequency
can allow for a more in depth interpretation of seismic data,
as both a thin layer detection method and a potential DHI.
Conclusions
Figure 3. Sand-2 horizon amplitude extraction in the far stack
showing an arbitrary line corresponding to the vertical display in Phase decomposition provides a new point of view for
Figure 4. interpretation of seismic data, allowing decomposition of
amplitude according to phase. Detecting thin layers (n a
As observed in Figure 4a, Well-1 intercepts 2 clear bright simpler fashion, when compared to inversion), porosity
spots that correspond to sands 1 and 2. Well-2 also crosses changes, and presence of hydrocarbons are some of the
Sand-2 in weaker but still anomalous amplitude. The phase benefits that can be achieved by studying this attribute.
decomposition results for the full bandwidth seismic are There is an evident relationship between amplitudes,
observed in Figures 4b (even) and 4c (odd). Note that the frequency, and phase that needs to be carefully studied to
amplitude anomalies move to the even component. This is efficaciously interpret the different phase components.
consistent with what the synthetics show, as we are Well information is not utilized in the phase
resolving these target sands, neglecting the thin bed decomposition; however, synthetic models can help the
requirement. The phase for isolated reflectors at top and interpreter to better understand the relationship between
base of Sand 2 thus have the even phase of the wavelet. amplitude and phase as a function of frequency. After
Figure 4d shows the same input data after a high cut establishing the required frequency content necessary to
bandpass filter. Note that the seismic section has not observe the separation of amplitudes between the different
changed drastically and that the amplitude anomalies phase components, seismic ‘thinning’ can be achieved by
remain. Figures 4e (even) and 4f (odd) show the phase high cut band pass filtering the data. In the case of the
decomposition result in the bandpassed stack. The studied Gulf of Mexico 3D seismic far stack we observe
difference is clear. After effectively seismically ‘thinning’ that the low impedance light oil bearing sands are separated
these layers, we observe the response predicted from the from brine bearing sands (of similar thicknesses) when
synthetics. Well-1 crosses 2 clear bright spots in the odd comparing phase components. This can be a powerful tool
component (which are hydrocarbon bearing) and the in order to de-risk prospects, establish drilling priorities,
original bright spot crossed by Well-2 has its amplitudes and unmask anomalies that could be hidden when all
divided between even and odd components, rendering it no phases are stacked together.
longer anomalous. Given that we observe similar
thicknesses of the target sands from one well to the next,
Acknowledgments
the difference in phase response could potentially be
attributed to a hydrocarbon effect. If so, the odd component The authors would like to thank Talos Energy,
after bandpass shows additional potential both downdip for WestenGeco, and Lumina Geophysical, specially Emil
Sand 2 as well as at a younger stratigraphic level. Nassif, Carlos Moreno, Gabriel Gil, John Walker, and
Miguel Silva for providing the data set, software, and
Figure-5 shows a similar analysis applied to a horizon slice support for this project, as well as permission to publish
extraction for Sand-2. The stars represent the location this work.
where the wells cross the horizon. The red star is for Well-1
(pay) and the black for Well-2 (wet). Note that the Well-1
is placed at the edge of the West amplitude anomaly, while
Well-2 is also at the edge of the East anomaly (Figure 5a).
Note that after phase decomposition (Figures 5b and 5c),
the amplitudes are observed in the even component. Under
the current analysis, these could have been considered
© 2017 SEG Page 3978

Figure 4. Arbitrary line across wells 1 and 2 . Panels A through C show the results at original frequency bandwidth data, and D through F at
bandpass filtered data. Panel E (even) and F (odd) show the amplitudes separated between even and odd components, with the pay well (1)
crossing clear bright spots in the odd component.
Figure 5. Map extraction for Sand-2. The stars show the well crossings with respect to the horizon, where red is pay and black is wet. We
can observe that the anomaly in Well-1 is maintained in the odd component (de-risking it) and that the amplitudes related to the crossing of
Well-2 are divided between even and odd components, no longer anomalous. Map F correlates to the known pay occurrence in both wells.
© 2017 SEG Page 3979

EDITED REFERENCES
Note: This reference list is a copyedited version of the reference list submitted by the author. Reference lists for the 2017
SEG Technical Program Expanded Abstracts have been copyedited so that references provided with the online
metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web.
REFERENCES
Castagna, J. P., and S. Sun, 2006, Comparison of spectral decomposition methods: First Break, 24, 75–79.
Castagna, J. P., A. Oyem, O. Portniaguine, and U. Aikulola, 2016, Phase decomposition: Interpretation, 4,
SN1–SN10, http://doi.org/10.1190/INT-2015-0150.1.
Chakraborty, A., and D. Okaya, 1995, Frequency-time decomposition of seismic data using wavelet
based-methods: Geophysics, 60, no. 6, 1906–1916, http://doi.org/10.1190/1.1443922.
Meza, R., G. Haughey, J. P. Castagna, U. Barbato, and O. Portniaguine, 2016, Phase decomposition as a
hydrocarbon indicator: A case study: 86th Annual International Meeting, SEG, Expanded
Abstracts, 1839–1943, http://doi.org/10.1190/segam2016-13871199.1.
© 2017 SEG Page 3980

Microseismic event location with direct and reflected waves: A Woodford case study
Nidhal Belayouni, David A. Katz (Magnitude),
Vladimir Grechka, and Pete Christianson (Marathon Oil Company)
Summary ledge and Soma, 2013; Belayouni et al., 2015), the prac-
Obtaining precise hypocenters of microseismic events is tical challenges in using reflections lay in ability of a geo-
the primary objective of contemporary microseismic sur- physicist to identify reflected waves and associate them
veys. The sought precision usually hinges on two fac- with particular reflectors in or around a reservoir. In
tors: an accurate velocity model, its inaccuracy biasing addition, those reflectors have to be sufficiently strong
hypocenters of an entire event population, and an ade- and consistent across the area of interest to generate
quate data-recording aperture, its deficiency blurring the interpretable reflections. The presence of such reflec-
hypocenters of individual events and causing noisy ap- tors is obviously field-dependent, and their variable qual-
pearance of event population computed even in a highly ity could, perhaps, explain why Rutledge and Soma
precise velocity model. Our paper explores the possibil- (2013) were able to pick reflections and relocate less than
ity of aperture enhancement through adding reflections to 4% of events comprising their microseismic catalogue,
direct arrivals, conventionally used to locate microseismic whereas Belayouni et al. (2015) illustrated the reduc-
events. tion of hypocenter-location uncertainty on just two events
(their Figures 9 through 12). On the other hand, Daniel
We illustrate the performance of our multiphase (that is, et al. (2016), monitoring production-related seismicity in
direct + reflected waves) event-location technique on a the Groningen gas field, The Netherlands, managed to
data set recorded in the Anadarko Basin Woodford play, locate approximately half of their catalogue with a com-
Oklahoma, USA, and demonstrate that the use of re- bination of direct, reflected, and refracted waves, demon-
flected waves not only improves velocity model but also strating that most of their events were triggered within
unambiguously places the recorded events in the Wood- the reservoir.
ford formation, something that cannot be achieved with
the direct arrivals alone. We have a similar task, although in conjunction with
hydraulic stimulation rather than gas production. In
Introduction our microseismic survey, acquired in the Anadarko Basin
Woodford play, Oklahoma, USA, a single array of three-
The precision of hypocenters, ξ ≡ [ξ1 , ξ2 , ξ3 ], of mic- component sensors was deployed above the Woodford for-
roseismic events located in hydraulic-fracturing applica- mation (see Figure 2 below), making it difficult to de-
tions is known to strongly depend on the geometry em- cide whether the recorded seismicity was confined to the
ployed to collect microseismic data. For example, sur- Woodford or also occurring in the adjacent formations.
face data-acquisition geometries can accurately position Including reflections from the interface right beneath the
the hypocenters laterally (in a laterally unbiased veloc- Woodford in our event-location procedure helped us re-
ity model), whereas the hypocenter depths ξ3 estimated solve the ambiguity and confidently state that only the
in surface microseismic are usually less precise than lat- reservoir itself has been stimulated.
eral components ξ1 and ξ2 of the hypocenters. Downhole
data-recording geometries, more diverse than surface ge- We begin our paper with analyzing a synthetic data set to
ometries, typically require placing receiver arrays in sev- gain an insight into the origin of improvements in the res-
eral observation wells (e.g., Figure 1 in Dohmen et al., olution of hypocenter depths brought by reflected waves
2014) to ensure the precision of ξ. Nonetheless, single- and then focus on the field data.
well monitoring geometries are still popular in the indus-
try despite their inability to provide equally high precision Synthetic data example
of all three hypocenter coordinates ξ1 , ξ2 , and ξ3 . In preparation for the forthcoming field case study, we
A particular single-well geometry yielding precise event examine the influence of reflected waves on the preci-
depths contains a sufficiently long receiver array strad- sion of event locations. To this end, we generate ray-
dling the formations in which hypocenters are expected; traced P- and SV-wave traveltimes between hypocenters
then the event depths ξ3 are close to the depths of the of two events and receivers placed in a single vertical mon-
moveout minima observed on microseismic records, mak- itor well in a horizontally layered vertically transversely
ing the computation of ξ3 almost velocity-independent. isotropic (VTI) model described in Table 1 and shown in
Conversely, a receiver array deployed above a reservoir Figure 1.
(usually to avoid drilling a dedicated monitor well) en- The rotational symmetry of the model around the vertical
tails a banana-shaped event-location uncertainty volume makes event azimuths unconstrained by the traveltimes,
that might cover a sizable range of depths (Figure 15 necessitating the use of the P-wave polarization azimuths
in Grechka and Duchkov, 2011, or Figure 13 in Grechka α for computing the hypocenters ξ. Substituting the po-
et al., 2011) and lead to poor estimates of ξ3 . Here, we larization azimuth, usually estimated from the hodogram
focus on such data-recording geometries and show that analysis of recorded particle motion, for the event azimuth
complementing direct-wave arrivals with reflections from simplifies the event-location problem, yielding the rela-
interfaces beneath the reservoir can constrain the event tionships ξ1 = % cos α and ξ2 = % sin α (% being the radial
depths to a level acceptable and useful in practice. distance between the event hypocenter and the observa-
Although the value of reflected waves for improving the tion well) that reduce three hypocenter-related unknowns,
precision of hypocenter depths has been recognized (Rut- ξ = [ξ1 , ξ2 , ξ3 ], to two, ξ = [%, ξ3 ], and make it possible
© 2017 SEG Page 3981

Microseismic event location with direct and reflected waves
(a)
Layer Depth of VP0 VS0 δ 4.1
bottom (km/s) (km/s) 4.15
(km)
4.2
1 4.30 4.0 2.2 0.10 0.05
4.25
2 4.45 5.0 2.9 0.05 0.03
4.3
Depth (km)
3 4.60 3.0 1.6 0.35 0.20
4.35
Table 1: Thomsen (1986) parameters of VTI model for syn-
thetic-data example in Figure 1. 4.4
4.45
to compute the hypocenters in the vertical [%, ξ3 ] plane 4.5
displayed in Figure 1.
4.55
We contaminate our traveltime data with the zero-mean 1 2
Gaussian noise that has (quite large for downhole mic- 4.6
roseismic) standard deviation of 2 ms and, following 0 0.1 0.2 0.3 0.4 0.5 0.6
Grechka and Yaskevich (2013, 2014), apply the joint mic- Radial distance (km)
roseismic tomography to compute the hypocenters of two (b)
events (the large blue dots) in Figure 1 simultaneously 4.1
with Thomsen (1986) parameters listed in Table 1. The
event hypocenters obtained with the direct P- and SV- 4.15
wave data for 100 random realizations of the noise are 4.2
presented in Figure 1a (the cyan dots). We clearly see
the event populations spreading in the directions approx- 4.25
imately orthogonal to the directions of rays (black and 4.3
gray) at the sources, reminiscent of field-data observa- Depth (km)
tions reported in Grechka et al. (2011). The standard 4.35
deviations of the event depths are 4.4

[1] [2] 4.45
std ξ3 = 14.7 m and std ξ3 = 23.6 m. (1)
4.5
Adding the reflected P-waves to the direct waves (Fig- 4.55

ure 1b) profoundly changes the shapes of the hypocenter 1 2
4.6
populations, their depth distributions now characterized
0 0.1 0.2 0.3 0.4 0.5 0.6
by the standard deviations Radial distance (km)

[1]

[2] Fig. 1: Hypocenters of microseismic events (the cyan dots)
std ξ3 = 5.9 m and std ξ3 = 9.8 m, (2) computed from noise-contaminated traveltimes of (a) direct
P- and SV-waves and (b) direct P-, SV-waves and reflected P-
a marked reduction compared to equation 1. waves simultaneously with estimating a model containing three
horizontal VTI layers. The rays of the direct P-, SV-waves
The ray trajectories in Figure 1b provide an intuitive ex- and the reflected P-waves are shown in black, gray, and red,
planation of why the higher precision has been achieved. respectively. The black triangles indicate receivers in a single
Indeed, moving the events, for example, down would in- vertical well, the large blue dots are the true event hypocenters.
crease the direct-wave times and decrease the reflected-
wave times in our geometry for any fixed values of other
three-component (3C) receives deployed in a single mon-
model parameters. As there is no parameter in the model
itor well shown in Figure 2 (two receivers stopped work-
space that would compensate for such opposite traveltime
[1] [2] ing soon after deployment, leaving 18 active receivers).
changes, the hypocenter depths ξ3 and ξ3 are well con- The main objective of this particular microseismic survey
strained; the same logic applies when the events move up was to find out whether hydraulic treatment of a hori-
rather than down. Clearly, the depth constraints emerge zontal well drilled in approximately north-south direction
because of the presence of both up- and downgoing rays in the Woodford formation (the black stars in Figure 2)
in the bottom layer in Figure 1b; the later, related to the efficiently stimulates the Woodford shale reservoir or it
reflected arrivals (the red rays in Figure 1a), entail sub- also fractures the adjacent Caney, Sycamore, and Hunton
stantial increase of the polar ray aperture in the bottom formations.
layer compared to the aperture provided by the direct
rays only (the black and gray rays in Figure 1a and 1b). The standard microseismic data processing, relying ex-
clusively on the direct P- and S-wave arrivals, has proven
incapable of meeting our objective. It produced a popu-
Woodford case study lation of events similar to that shown with the blue dots
Conventional processing in Figure 2 that extend by over 100 m into the formations
above and below the Woodford.
Having understood the constraints on event depths im-
posed by reflected waves, we turn our attention to a data When processing the data, we discovered 9 perforation
set recorded in the Woodford play with an array of 20 shots (the stars in Figure 2) and 80 microseismic events
–2–
© 2017 SEG Page 3982
(a)
4.2 2
4
4.3 6
Depth (km)
Springer
Sensor number
8
4.4 Caney 10
Sycamore 12
4.5 Woodford 14
16
Hunton
4.6 18
20
-0.6 0.25 0.3 0.35 0.4 0.45 0.5

-0.4 -0.2 0 0 0.2 0.4
Time (s)
Local East (km) Local North (km)
(b)
Fig. 2: Side view of data-acquisition geometry and events lo-
cated in our field data set. The receivers, placed in a single 2
nearly vertical monitor well, are marked with the black trian- 4
gles, the perforation shots – with the black stars. The gray sur-
6
faces appearing as thick lines are the planar model interfaces,
Sensor number
dipping at 2.5◦ from northeast to southwest. The hypocen- 8
ters of 80 events computed using the direct P- and S-waves 10
only are indicated by the blue dots, the hypocenters of the 12
same events computed using the direct P-, S-waves and the P-
14
waves reflected from the Woodford-Hunton interface – by the
red dots. 16
18
20
exhibiting distinct arrivals following the direct P-waves
0.25 0.3 0.35 0.4 0.45
(Figure 3). After full waveform modeling (not shown) Time (s)
conducted in a log-derived seismic model, we identified
those arrivals as the P-wave reflections at the strong Fig. 3: Records of representative (a) perforation shot and (b)
microseismic event containing recognizable direct and reflected
Woodford-Hunton interface (see the sonic in Figure 4a). P-waves. The time picks of the P- and S-waves are shown with
the green dots; the red components of the three-component
Anisotropic velocity-model building traces point to the east, gray – to the north, and blue – down.
To locate events using both direct and reflected waves,
we construct a model containing 11 dipping layers (Fig- (a) (b)
4.1 4.1
ures 2 and 4), their dips, 2.5◦ from northeast to southwest, ǫ(1)
matching those in seismic reflection data and the dip of ǫ(2)
the treatment well, their depths tied to sonic logs. We 4.15 4.15 δ(1)
extend the joint microseismic velocity-estimation/event- δ(2)
location tomography (Grechka and Yaskevich, 2013, 2014) δ(3)
4.2 4.2
to include reflected-wave traveltimes and obtain the in- γ(1)
terval vertical velocities and anisotropy parameters pre- γ(2)
sented in Figures 4. The top 9 model layers are con- 4.25 4.25
strained to VTI, the Woodford layer is assumed to be
orthorhombic with a horizontal symmetry plane, and the
bottom Hunton layer is isotropic, its P- and S-wave ve- 4.3 4.3
Depth (km)
Depth (km)
locities are extracted from the sonic logs in Figure 4a. Springer
The presence of both up- and downgoing rays in the 4.35 4.35
Woodford (the red dots in Figure 5) turns out to be instru-

mental for constraining its orthorhombic symmetry and Caney
4.4 4.4
quantifying azimuthal anisotropy (in terms of Thomsen-
Sycamore
style coefficients defined in Tsvankin, 1997) of the Wood-
ford formation. Quite interestingly, the azimuth E2.9◦ N 4.45 4.45
of the vertical [x2 , x3 ] symmetry plane, approximately
Woodford
orthogonal to the direction of the horizontal treatment
well, and the greater Tsvankin’s coefficients (1) , δ (1) , γ (1) 4.5 4.5
in that plane (shown with the black, blue, and magenta

Hunton
lines in Figure 4b) than their counterparts (2) , δ (2) , γ (2) 4.55 4.55
1 2 3 4 5 6 0 0.1 0.2 0.3
in the [x1 , x3 ] plane (the red, cyan, and gray lines in Fig- Velocity (km/s) Anisotropy coefficient
ure 4b), approximately along the wellbore, suggest that
Fig. 4: (a) Sonic logs (thin blue and black), vertical velocities
the triplet of coefficients (1) , δ (1) , and γ (1) is represen- (blocky cyan and blocky gray), and (b) anisotropy coefficients
tative of the anisotropy of unfractured Woodford shales, (Tsvankin, 1997). The vertical velocities and anisotropy coef-
whereas the triplet (2) , δ (2) , and γ (2) bears the impact ficients are estimated in a layered model used to locate micro-
of the treatment-induced fractures, extending approxi- seismic events in Figure 2.
–3–
© 2017 SEG Page 3983
the direct waves only are not confined to the Woodford

formation (the blue dots in Figure 2) despite being com-
puted in exactly the same anisotropic velocity model, the
differences between the two event populations – red and
blue in Figure 2 – illustrating the influence of the aper-
ture at the event sources (compare the red and blue dots
in Figure 5) on the propagation of noise into the estimated
hypocenters.
Discussion and conclusions

Apart from reaching our practical objective and establish-
ing the containment of all analyzed microseismic events
within the Woodford formation, our paper presents two
technical developments:
(1) extension of the joint microseismic velocity-estima-
tion/event-location tomography to reflected waves,
and
Fig. 5: Directions of rays in the Woodford formation plotted (2) construction of the first to the best of our knowledge
on the surface of the unit sphere. The rays corresponding to orthorhombic model of the Woodford shale forma-
the blue and red event hypocenters in Figure 2 are shown in tion.
blue and red, respectively. We discuss those two developments in turn.
Including reflections in the tomographic approach of
4
Grechka and Yaskevich (2013, 2014) is operationally
straightforward: one just needs to trace the reflected rays
3
and compute the Fréchet derivatives of the reflected-wave
times with respect to relevant model parameters. Prac-
2 tically though, the requirement of a high accuracy of the
Traveltime misfit (ms)
reflector depth and dip poses immediate challenge, the

1 challenge that does not exist for direct arrivals, for which
errors in the placement of model interfaces can be largely
0 compensated by varying velocities and anisotropy coeffi-
cients in the model layers. With traveltimes of the reflec-
-1 tions being added, however, the depths of events above
the reflector are largely controlled by the observed time
-2 delays between the direct and reflected waves (Figure 3),
making the hypocenter depths inherit any inaccuracy in
-3 the reflector positioning.
The contribution of reflected-wave data to the estimation
-4 of anisotropy of the Woodford formation is well illustrated
1000 2000 3000 4000 5000
Traveltime pick in Figure 5, exhibiting the differences in the polar aper-
tures of the up- and downgoing rays (the red dots in the
Fig. 6: Traveltime misfits for direct and reflected waves excited
upper and lower hemispheres in Figure 5) in addition to
by perforation shots (the black stars in Figure 2) and micro-
seismic events (the red dots in Figure 2) in the velocity model the obvious difference in the signs of the corresponding
displayed in Figures 2 and 4. polar angles. Clearly, combining the apertures of both the
direct and reflected arrivals increases the total aperture,
helping constrain anisotropy parameters of the Woodford.
mately normally to the well trajectory. Another important feature of our anisotropic velocity
As usual, the quality of velocity model constructed to pro- model used for the computation of hypocenters in Fig-
cess microseismic data can be judged by the magnitude ure 2 is the absence of its rotational symmetry because
of misfits between the times picked from recorded data of both nonzero interface dip and azimuthal anisotropy in
(the green dots in Figure 3) and the times computed in the Woodford. As a result, unlike our synthetic data ex-
the derived model. The individual misfits displayed in ample, the P-wave polarization azimuths at the receivers
Figure 6 yield the root-mean-square misfit of 0.86 ms for are no longer equal to the event azimuths. Their dif-
the collection of 9 perforation shots and 80 microseismic ferences entail the differences of the azimuths of the red
events (on data sampled at 0.5 ms), attesting to the high and blue rays, observable in the upper hemisphere in Fig-
quality of our velocity model. ure 5, and confirm that maintaining the recorded P-wave
polarization azimuths necessarily couples the vertical and
Event locations lateral movements of the event hypocenters.
Microseismic events located in the model from Figure 4 Finally, we mention one possible extension of the pre-
with both direct and reflected waves line up in the Wood- sented work. Accurate hypocenters of other microseismic
ford formation, the microseismic cloud growing up from events that were recorded in our data set but did not
perforations in the treatment well and propagating to- generate pickable reflections can be computed relatively
wards the top of the Woodford (the red dots in Figure 2). to the already located hypocenters (the red dots in Fig-
Contrary to that, the hypocenters of events located with ure 2), as discussed in Grechka et al. (2015, 2016).
–4–
© 2017 SEG Page 3984
EDITED REFERENCES
REFERENCES
Aki, K., and P. G. Richards, 1980, Quantitative seismology: W. N. Freeman & Company (2nd ed., 2002,
University Science Books).
Baig, A., and T. Urbancic, 2010, Microseismic moment tensors: A path to understanding frac growth: The
Leading Edge, 29, 320–324, https://doi.org/10.1190/1.3353729.
Ben-Menahem, A., and S. J. Singh, 1981, Seismic waves and sources: Springer Verlag (2nd ed., 2000,
Dover Publications).
Dattorro, J., 2005, Convex optimization & Euclidean distance geometry: Meboo Publishing.
Diller, D. E., T. Shuck, and B. Fish, 2015, Estimation and interpretation of high-confidence microseismic
source mechanics: The Leading Edge, 34, 918–924, https://doi.org/10.1190/tle34080918.1.
Dohmen, T., J. P. Blangy, and J. Zhang, 2014, Microseismic depletion delineation: Interpretation, 2,
SG1–SG13, https://doi.org/10.1190/INT-2013-0164.1.
Dufumier, H., and L. Rivera, 1997, On the resolution of the isotropic component in moment tensor
inversion: Geophysical Journal International, 131, 595–606, https://doi.org/10.1111/j.1365-
246X.1997.tb06601.x.
Eaton, D. W., and F. Forouhideh, 2011, Solid angles and the impact of receiver-array geometry on
microseismic moment-tensor inversion: Geophysics, 76, no. 6, WC77–WC85,
https://doi.org/10.1190/geo2011-0077.1.
Eisner, L., S. Williams-Stroud, A. Hill, P. Duncan, and M. Thornton, 2010, Beyond the dots in the box:
Microseismicity-constrained fracture models for reservoir simulation: The Leading Edge, 29,
326–333, https://doi.org/10.1190/1.3353730.
Grechka, V., Z. Li, and B. Howell, 2016, Relative location of microseismic events with multiple masters:
Geophysics, 81, no. 4, KS149–KS158, https://doi.org/10.1190/geo2015-0445.1.
Grechka, V., and S. Yaskevich, 2013, Inversion of microseismic data for triclinic velocity models:
Geophysical Prospecting, 61, 1159–1170, https://doi.org/10.1111/1365-2478.12042.
Grechka, V., and S. Yaskevich, 2014, Azimuthal anisotropy in microseismic monitoring: A Bakken case
study: Geophysics, 79, no. 1, KS1–KS12, https://doi.org/10.1190/geo2013-0211.1.
Hudson, J. A., R. G. Pearce, and R. M. Rogers, 1989, Source type plot for inversion of the moment
tensor: Journal of Geophysical Research, 94, 765–774,
https://doi.org/10.1029/JB094iB01p00765.
Jechumtálová, Z., and L. Eisner, 2008, Seismic source mechanism inversion from a linear array of
receivers reveals non-double-couple seismic events induced by hydraulic fracturing in
sedimentary formation: Tectonophysics, 460, 124–133,
https://doi.org/10.1016/j.tecto.2008.07.011.
Kuehn, D., H. N. Gharti, V. Oye, and M. Roth, 2009, Automatic determination of full moment tensor
solutions from P-wave first motion amplitudes: 71th Annual International Conference and
Exhibition, EAGE, Extended Abstracts, A05, https://doi.org/10.3997/2214-4609.20146731.
Lee, M., T. Davis, and S. Maxwell, 2014, The use of amplitude ratios to constrain source mechanisms of
microseismic data: A case study from the Montney Shale: First Break, 32, 67–73.
Li, J., C. Li, S. A. Morton, T. Dohmen, K. Katahara, and M. N. Toksöz, 2014, Microseismic joint location
and anisotropic velocity inversion for hydraulic fracturing in a tight Bakken reservoir:
Geophysics, 79, no. 5, C111–C122, https://doi.org/10.1190/geo2013-0345.1.
© 2017 SEG Page 3985

Nolen-Hoeksema, R. C., and L. J. Ru , 2001, Moment tensor inversion of microseisms from the B-sand
propped hydrofracture, M-site, Colorado: Tectonophysics, 336, 163–181,
https://doi.org/10.1016/S0040-1951(01)00100-7.
Rodriguez, I. V., Y. J. Gu, and M. D. Sacchi, 2011, Resolution of seismic-moment tensor inversions from
a single array of receivers: Bulletin of the Seismological Society of America, 101, 2634–2642.
Rutledge, J., R. Downie, S. Maxwell, J. Drew, and T. Fischer, 2013, Extension-shear microseismic
mechanisms during hydraulic fracturing: 83rd Annual International Meeting, SEG, Expanded
Abstracts, 2067–2072, https://doi.org/10.1190/segam2013-1387.1.
Rutledge, J., and W. S. Phillips, 2003, Hydraulic stimulation of natural fractures as revealed by induced
microearthquakes, Carthage Cotton Valley gas field, east Texas: Geophysics, 68, 441–452,
https://doi.org/10.1190/1.1567214.
Shearer, P. M., 2009, Introduction to seismology: Cambridge University Press.
Šílený, J., L. Eisner, D. P. Hill, and F. H. Cornet, 2009, Non-double-couple mechanisms of
microearthquakes induced by hydraulic fracturing: Journal of Geophysical Research, 114,
B08307, https://doi.org/10.1029/2008JB005987.
Song, F., and M. N. Toksöz, 2011, Full-waveform based complete moment tensor inversion and source
parameter estimation from downhole microseismic data for hydrofracture monitoring:
Geophysics, 76, no. 6, WC103–WC116, https://doi.org/10.1190/geo2011-0027.1.
Stanek, F., and L. Eisner, 2013, New model explaining inverted source mechanisms of microseismic
events induced by hydraulic fracturing: 83rd Annual International Meeting, SEG, Expanded
Vavryčuk, V., 2005, Focal mechanisms in anisotropic media: Geophysical Journal International, 161,
334–346, https://doi.org/10.1111/j.1365-246X.2005.02585.x.
Vavryčuk, V., 2007, On the retrieval of moment tensors from borehole data: Geophysical Prospecting, 55,
381–391, https://doi.org/10.1111/j.1365-2478.2007.00624.x.
Vavryčuk, V., 2011, Tensile earthquakes: Theory, modeling, and inversion: Journal of Geophysical
Research, 116, B12320, https://doi.org/10.1029/2011JB008770.
Vavryčuk, V., 2015, Inversion for the composite moment tensor: Bulletin of the Seismological Society of
America, 105, 3024–3035, https://doi.org/10.1785/0120150163.
Yang, Y., M. Zoback, M. Simon, and T. Dohmen, 2013, An integrated geomechanical and microseismic
study of multi-well hydraulic fracture stimulation in the Bakken formation: Unconventional
Resources Technology Conference, URTeC 1580301, 534–543,
https://doi.org/10.1190/urtec2013-056.
Yang, Y., and M. D. Zoback, 2014, The role of preexisting fractures and faults during multistage
hydraulic fracturing in the Bakken Formation: Interpretation, 2, SG25–SG39,
https://doi.org/10.1190/INT-2013-0158.1.
Yu, X., J. Rutledge, and S. Leaney, 2015, Event grouping and multi-event moment-tensor inversion for
ill-posed monitoring geometry: 85th Annual International Meeting, SEG, Expanded Abstracts,
2522–2526, https://doi.org/10.1190/segam2015-5876354.1.
© 2017 SEG Page 3986

Permian Basin Seismic Data Reprocessing – A Case Study
Yan Yan, Xianhuai Zhu, Junru Jiao, Pan Deng, Bin Yang and Lei Fu, Forland Geophysical Services (FGS)
Ernesto V. Oropeza, Jianxiong Chen and Sarah Mueller, Anadarko Petroleum Corporation
quality for better interpretation, fault definition, and sweet

spot identification at the target zone interval.
NEAR-SURFACE MODEL AND DE-MULTIPLES
An extensive effort was made in de-multiple which

involved incorporating the near-surface velocity depth
model derived from turning-ray tomography. In this study,
the multiples do not appear as lower velocity events on
gathers and the traditional techniques (FK, radon, stack,
etc) that rely on velocity discrimination to remove multiples
are ineffective. Predictive deconvolution has proven to be a
useful tool for the suppression of inter-bed multiples. The
key parameters are the prediction distance and operator
length. The prediction distance is generally set equal to the
multiple periods, while the operator length is often set
approximately equal to the wavelet length (Yilmaz, 1987).
A high-resolution near-surface velocity model was derived
Figure 1: Location map of the study area from a proprietary turning-ray tomography algorithm (Zhu,
Source: http://www.vyey.com/assets/permian-basin 1992), which was used to identify the generators of inter-
bed multiples and guide us in determining the multiple
periods. The turning-ray tomography uses first-break
SUMMARY traveltimes and locations of sources and receivers to
estimate the near surface velocity models.
This paper presents a case study of reprocessing seismic
data in the Permian Basin. The legacy data was previously In Figure 2, the distance between two high-velocity layers
processed in 1997. Due to noise, including multiples in the shallow section, being the generator of inter-bed
associated with evaporites and salt near the surface, the multiples, was 220ft. The velocity between the two layers
legacy results were difficult to interpret. was 14500ft/s. This resulted in approximately a 30 ms two-
way traveltime as the multiple period.
Reprocessing was completed recently using an integrated
approach. With high-resolution turning-ray tomography,
adaptive noise removal, 5D interpolation and post-stack
structure-oriented filtering algorithms, we improved the
seismic images, resulting in better signal-to-noise ratio
(SNR), fewer inter-bed multiples, and less surface related
noise. This is significant for interpretation and horizontal
drilling in the Permian Basin.
INTRODUCTION
The study area is located in the Delaware Basin (Figure 1),

which is a part of the Permian Basin in the USA. The
previously processed data, or legacy data, has a low SNR
and the images were challenging to interpret at the target
zone. This is because of the near-surface complexities, such
as evaporites and a velocity reversal, resulting in inter-bed
multiples, surface waves, and scattering noise. The purpose
of the reprocessing was to improve the seismic image Figure 2. Near-surface velocity model by turning-ray
tomography
© 2017 SEG Page 3987

Permian Basin Seismic Reprocessing
The multiples’ period (gap) can also be designed by high-energy surface wave within a window of traces using
examining the autocorrelations of traces prior to the eigenimage and then the reconstructed noise was
deconvolution. Figure 3 (left) shows the autocorrelation of subtracted (Carry and Zhang, 2009). Secondly, after
a shot with multiples, which also shows the predictive eigenvector filtering, we input two versions of processed
distance of an average of 30 ms. It matched well with the seismic data into signal/noise adaptive processing (SNAP):
gap derived from the near-surface velocity model by one version lightly processed and the other version heavily
turning-ray tomography. Figure 3 (right) shows the processed. Then we selectively replaced noisy portions in
autocorrelation of the same shot after applying predictive lightly processed data with the more heavily processed data
deconvolution with a gap of 30ms. Comparing these two - this process was implemented by comparing their trace
autocorrelations (Figure 3) and stack sections (Figure 4), difference with a chosen time-variant threshold. The
we can see that most interbed multiples were attenuated. assumption for this algorithm is that noise has higher
amplitude than signals so that we keep the signal portions
undamaged. This flow removed noise without introducing
undesirable artifacts in places where the noise does not
exist.
The shot gathers before (left) and after (right) de-noising

are shown in Figure 5. The difference is shown in Figure 6.
Reflections (around 2000 ms) start to show up in the shot
gathers after de-noise in Figure 5 (right).
Figure 3. Autocorrelation before (left) and after de-multiple (right)
Figure 5. The shot before (left) and after de-noising (right)
Figure 4. Stack section after tomo-refraction statics (a) followed by

gap decon (b). Interbed multiples are significantly reduced in (b).
ADAPTIVE NOISE REMOVAL
De-noising of the data is a challenging problem since the

seismic data are heavily contaminated by different noise
types including aliasing, surface waves, multiples, and
reverberations from shallow high-velocity layers. Figure 6. The difference before and after de-noising
A two-pass approach was used for de-noising in the study

area. Firstly, an eigenvector filter was used to estimate the
© 2017 SEG Page 3988

5D INTERPOLATION POST-STACK ENHANCEMENT

The 5D interpolation method is based on Fourier Structure-oriented filtering was applied to this data to
reconstruction by Minimum Weighted Norm Interpolation further clean-up and enhance the image to improve
(MWNI). It operates on 5 dimensions of the seismic data subsurface understanding – without remigration. Structure
(Inline, Xline, Offset, Azimuth and Temporal). In this
tensors (Hale, 2009) were calculated following the
study, it increased the cdp fold (Figure 7), reduced noise
and proved to be a useful tool to precondition data before orientation of the structure, linear or point like features,
the subsequent velocity analysis (Figure 8 and 9). then an edge-preserving filter was used.
Figure 7 (left) shows the original cdp fold map with Figure 10 shows the PSTM stack sections before (left) and
maximum fold of 51, and Figure 7 (right) shows that cdp after (right) structure-oriented filter.
fold is about 4 times higher with maximum fold of 189 after
5D interpolation.
Figure 8 shows a cdp gather with NMO before (left) and

after (right) interpolation. It clearly can be seen that the
gather after interpolation provided much more information
for the velocity analysis. Note that 5D interpolation has
been proved to be useful only after noise removal.
Figure 9 shows the stack section before (left) and after

(right) interpolation. The interpolated section has better
SNR and continuity.
Figure 9. Stack before (left) and after (right) interpolation.
Figure 7. CDP fold map before (left) and after (right) interpolation
Figure 10. PSTM before (left) and after (right) structure-oriented filter
(ZOI stands for Zone of Interest).
PSTM COMPARISONS
The legacy data, the PSTM stacks previously processed by

the third party in 1997, are shown in Figure 11 (left) and 12
(left), and the recently re-processed final PSTM results are
shown in Figure 11 (right) and 12 (right). The re-processed
Figure 8. CDP gather before (left) and after (right) interpolation results exhibit an improved SNR and better imaging of the
target interval and deep faults with less noise and fewer
multiples.
© 2017 SEG Page 3989

Figure 13 shows a time slice of PSTM stack at 2000ms. CONCLUSIONS

The legacy data is shown on the left and current processing
result is shown on the right. Clearly the re-processed result The recently re-processed results show improved images,
shows a cleaner image with better focusing. which indicate the effectiveness of applying the latest
processing techniques. The near-surface velocity depth
model derived from turning-ray tomography is the key for an
accurate identification of multiple source and multiple
period to remove interbed multiple successfully. The de-
noise with adaptive noise removal algorithm improved event
continuity and fault definition. More accurate velocity
analysis benefited from 5D interpolation that overcame the
acquisition constraints to yield higher folds and cleaner data.
Post-stack structure-oriented filtering was also implemented
to improve the SNR.
The reprocessing has improved seismic images for

interpretation and subsequent horizontal drilling, indicating
Figure 11. Inline PSTM Stack – Legacy (left) & Current (right) that legacy data can be utilized to add values to
unconventional resource play areas such as the Permian
Basin.
ACKNOWLEDGMENTS
We thank Anadarko Petroleum Corporation for the

opportunity of reprocessing this dataset and the show rights.
Constructive discussions and suggestions from David Sixta,
Tim Fasnacht and Allen Bertagne have greatly improved
the reprocessing results. The seismic data were acquired by
WesternGeco. We also thank FGS for permission to present
this work.
Figure 12. Xline PSTM Stack – Legacy (left) & Current (right)
Figure 13. Depth slice of PSTM Stack – Legacy (left) & Current (right)
© 2017 SEG Page 3990

EDITED REFERENCES
REFERENCES
Cary, P. W. and C. Zhang, 2009, Ground roll attenuation with adaptive eigenimage filtering: 79th Annual
International Meeting, SEG, Expanded Abstracts, 3302–3306, https://doi.org/10.1190/1.3255545.
Hale, D., 2009, Structure-oriented smoothing and semblance: CWP Report 635.
Yilmaz. O. 1987, Seismic data processing: SEG.
Zhu, X., D. P. Sixta, and B. G. Angstman, 1992, Tomostatics: Turning-ray tomography + static
corrections: The Leading Edge, 11, 15–23, https://doi.org/10.1190/1.1436864.
© 2017 SEG Page 3991

Correlation of azimuthal velocity anisotropy and seismic inversion attributes to Austin Chalk
production: a south central Texas case study
Keller, William, Rene´ Mott, Austin Jumper, EnerVest, Ltd.; Heloise Lynn, Walt Lynn, Lynn, Inc.; Mike Perz,
Arcis Seismic Solutions: A TGS Company
Summary greatly simplifies our analysis in comparison to other, more

recently developed unconventional plays.
The Austin Chalk play is a productive geologic trend that
extends from southern Maverick County, TX near the Within Giddings Field, the Austin Chalk formation ranges
Mexican Border more than 300 miles to the northeast to from 6,000-11,000 ft depth. Within the 3D seismic test
Madison County, TX. Giddings Field lies near the eastern area, the Austin Chalk is approximately 800 ft thick and is
end of this trend and has been a focus of Austin Chalk underlain by 150 ft of Eagle Ford shale. The highly
development since the early 1970s. In 2015, a 710 square fractured nature of the Austin Chalk and Eagle Ford Shale
mile, high-fold, wide-azimuth onshore 3D seismic survey as well as the large combined thickness of the interval
was acquired with a primary objective of imaging the make this an ideal candidate for VVAZ analysis. Indeed,
Austin Chalk and Eagle Ford reservoir intervals within prior authors have noted high degrees of azimuthal
Giddings Field. Azimuth-preserving, AVO-compliant anisotropy within the Austin Chalk formation in this area as
processing was performed over a 50 square mile test area well (Lynn, 2000).
within this dataset to evaluate the possible integration of
seismic inversion and azimuthal velocity anisotropy Methodology
(VVAZ) attributes to help guide future development of the
field. This evaluation revealed strong correlations between In 2015, Seitel, Inc. acquired a 195-fold wide-azimuth 3D
these seismic attributes and historical production within the that was designed with 990 ft source line spacing, 1155 ft
test area. receiver line spacing, 165 ft shot and receiver group
intervals, and an active patch size of 28,875 x 29,935 ft.
Introduction Natural binning for this survey is 82.5 x 82.5 ft.
The Upper Cretaceous Austin Chalk is a low permeability Within a 50 square mile subset of the 3D seismic survey,
formation made up of micritic chalks interbedded with processing was performed by Arcis Seismic Solutions
marls of varying clay content (Dawson, 1995 among using a flow designed to preserve amplitude variations
others). Average matrix porosity is generally very low, so across both offset and azimuth coordinates. One of the key
fracture systems with primary orientation parallel to the pieces of this processing flow involved 5D interpolation
underlying Lower Cretaceous shelf edge (NE-SW in the onto a mixed Cartesian-polar coordinate grid defined by
Giddings area) provide the permeability and storage cmp-x, cmp-y, offset, and azimuth. This choice of grid
capacity to support commercial oil and gas production definition allowed for the generation of a densely sampled
(Pearson, 2012). The source rock responsible for charging set of CMP gathers, regularly sampled across both offset
the fracture system within the Austin Chalk interval is the and azimuth domains. Interpolated traces that existed too
underlying organic rich Eagle Ford shale. far from recorded data traces to be considered reliable were
culled prior to grouping the remaining traces by common
Oil was first produced in commercial quantities from offset-azimuth indices to produce single-fold common
vertical wells within Giddings field beginning in 1973. A offset vector ensembles for input into prestack time
second phase of development coincided with the advent of migration (in this case a “VTI-aware” migration was
early horizontal drilling in the mid 1980’s but really came performed in which the travel time engine included the
into full swing in the early 1990’s as horizontal drilling effects of vertical transverse isotropy). Following VTI-
technology became more reliable and repeatable PSTM, the data were submitted to a post-migration noise
(Haymond, 1991). These wells were generally open-hole attenuation process operating piecewise on individual
completions that were treated with a small amount of acid azimuth “spokes”. This process is described in detail in a
before flow-back. Given this completion methodology, it companion paper (Perz et al., 2017). Finally these noise-
was vital for horizontal wells to encounter natural fractures attenuated data were input toVVAZ inversion via the
in order establish commercial production. It is not until Generalized Dix Inversion methodology in order to
more recently that wells have begun to be completed using generate Vint-fast, Vint-slow, and Vint-fast azimuth from
modern hydraulic fracturing techniques. As will be seen their corresponding RMS counterparts (Grechka, 1999).
later, this lack of hydraulic fracturing completion overprint
© 2017 SEG Page 3992

Correlation of azimuthal anisotropy and seismic inversion attributes to Austin Chalk production
Vint-slow values (red glyph colors). We interpret these

high percent anisotropy fairways to be associated with
Interpretation areas that exhibit a high degree of natural fracturing

oriented parallel to the Vint-fast azimuth. Imaging of fault
The resulting map from these interval VVAZ attributes for patterns within the test area suggests a highly complex
the Top Austin Chalk to Top Buda interval is shown in series of overprinting episodes of normal faulting at various
Figure 1. The map is comprised of glyphs plotted at the azimuths, likely related to movement of the underlying
center of 9 bin blocks throughout the 3D test area, the Louann Salt. However, it is interesting to note that despite
symbology of which is described in detail within the figure this structural complexity, Vint-fast azimuth within the
caption. Vint-fast is variable across the survey area, high anisotropy fairways is a very stable N45ºE. Perhaps
particularly in the vicinity of faults. However, areas with not coincidentally, this azimuth is parallel to the regional
high percent anisotropy (i.e. long glyphs) tend to have a maximum horizontal compressive stress, possibly
very stable Vint-fast azimuth of roughly N45ºE. These indicating that natural fractures with this orientation may
areas also tend to be associated with anomalously slow preferentially be more open than fractures oriented at other
Figure 1: Azimuthal interval velocity anisotropy map with cumulative production bubble overlay. The reservoir zone used for calculation of
the interval velocity attributes extends from the top of the Austin Chalk to the top of the Buda and thus encompasses the entirety of the Austin
Chalk and Eagle Ford formations. The interval velocity attributes are displayed as follows: glyph color = Vint-slow (warm colors are slower
than cold colors), glyph length = % Vint Anisotropy, glyph azimuth = Vint-fast azimuth. The map shows a clear visual correlation between the
interval velocity attributes and cumulative Austin Chalk production.
© 2017 SEG Page 3993

azimuths.
Interpretation of high percent anisotropy fairways as
corresponding to highly fractured areas is consistent with

the observation that these fairways are also associated with
elevated production (i.e. larger production bubbles shown
in Figure 1). Although not completely straightforward,
indications of pressure communication tend to occur
between producing wells along a roughly N45ºE azimuth
parallel to Vint-fast azimuth, and are most common within
areas that exhibit a high degree of fracturing. Indications of
pressure communication can include mud losses while
drilling infill wells, abrupt changes to offset well
production history once an infill well is brough online, etc.
The observation that pressure communication tends to
preferentially occur at an azimuth of N45ºE is also
generally consistent with the VVAZ attributes.
In addition to the strong visual correlation of cumulative

Austin Chalk production to the VVAZ attribute map in
Figure 1, a multivariate analytics model was generated
using Transform software and three seismic attributes as
input explanatory variables. In order to perform this
analysis, we first selected a subset of 50 horizontal Austin
Chalk producing wells that had reliable production history
Figure 3: a) Cumulative gas production grid generated from

Austin Chalk well production data. Warm colors indicate higher
Figure 2: Actual versus predicted cumulative gas production for 50 production. The white rectangle denotes the 3D seismic test area,
wells within the 3D seismic test area. Values were extracted from and well locations are shown in black. b) Seismic attribute
seismic attribute maps along horizontal wellbores and written to predicted cumulative gas production grid overlying the cumulative
wellbore zones. These zone values were then used as explanatory gas production grid from part a). Three seismic attributes were
input variables in a non-linear multivariate analytics model in order included as explanatory input variables in a non-linear multivariate
to predict cumulative gas production. The linear regression of actual analytics model in order to predict cumulative gas production.
vs. predicted values for cum production is ullustrated by the red line
above (correlation coefficient = 0.79, r-squared 0.63).
© 2017 SEG Page 3994

and were free of production allocation uncertainties. We Secondly, the Austin Chalk is a very thick and highly
then generated three seismic attribute maps for input into anisotropic interval that is ideally suited for this type of
the model – P Impedance, V-int slow, and Horizontal azimuthal velocity analysis. In cases where VVAZ results
Stress. P Impedance was generated using the prestack appear to be disappointing, the culprit may be lack of
simultaneous model based inversion methodology strong HTI anisotropy and or a target interval that is
described by Hampson and Russel (2005). V-int slow was insufficiently thick to create stable results.
generated using the VVAZ inversion methodology
described in the previous section of this paper, and The results generated with the three-term multivariate
Horizontal Stress was generated by utilizing the following analytics model are encouraging and appear to be
equation (Sayers, 2010): geologically reasonable within the test area. Future work
includes further refinement of this model and expansion of
Horizontal Stress = (PR)/(1-PR) * vertical stress this processing and interpretation methodology to a larger
area of the 3D survey.
In this case, poisson’s ratio (PR) is calculated from interval
velocities where Vp = Vint-slow and Vs is estimated from
Vint-slow using a linear transformation derived from Acknowledgements
measured dipole sonic logs within the test area.
The authors would like to thank Seitel, Inc. and EnerVest,
Values were extracted from these three seismic attribute Ltd. for permission to publish this paper as well as Arcis
maps along the horizontal wellbore trajectory for each of Seismic Solutions, A TGS Company, for carrying out the
the 50 wells to be included in the model. These values processing.
were then saved as well scalar attributes and used as input
variables for the non-linear multivariate analytics model.
The resulting actual cum gas production vs. predicted cum
gas production cross plot is shown in Figure 2. A linear
regression analysis of this cross plot yields a correlation
coefficient of 0.8 and an r-squared value of 0.63, a
remarkable result given the fact that no engineering or
drilling parameters were included in the analysis.
Once the multivariate analytics model was generated, we

were then able to calculate a production prediction map by
utilizing the three seismic attribute maps as input. The
resulting map is shown in Figure 3 b). One can easily see
the uplift in spatial resolution and granularity relative to the
gridded cumulative production map generated only from
well control that is shown in Figure 3 a). The features that
can be seen in the predicted production map look
geologically reasonable, and suggest that the fracture
fairways that are related to highest production can actually
be quite narrow and well defined linear features in map
view. Areas of lower production are likely related to areas
of less intense faulting and fracturing.
Conclusions
Seismic inversion and VVAZ attributes were generated

within a 50 square mile subset of a 3D seismic survey
acquired over Giddings Field. The VVAZ attributes
exhibit a strong visual correlation to cumulative Austin
Chalk production in the area and may be attributed to two
main factors. First, because of the history of development
within this field, the impact of completion overprint
appears to be minor. Most of the wells are open hole
completions that were not hydraulically fractured.
© 2017 SEG Page 3995

EDITED REFERENCES
REFERENCES
Dawson, W. C., K. Barry, and Robison, V. D., 1995, Austin Chalk petroleum system Upper Cretaceous,
southeastern Texas: A case study: Transactions — Gulf Coast Association of Geological
Societies, 45, 157–163.
Hampson, D. P., B. H. Russell, and B. Bankhead, 2005, Simultaneous inversion of pre-stack seismic data:
75th Annual International Meeting, SEG, Expanded Abstracts, 1633–1637,
http://doi.org/10.1190/1.2148008.
Haymond, D., 1991, The Austin Chalk — An overview: Houston Geological Society Bulletin, 33, 27–34.
Lynn, H. B., L. Weathers, and W. Beckham, 2000, The mobil onshore texas 3D full-azimuth full-offset P-
wave survey: 9th International Workshop on Seismic Anisotropy, Abstract at
http://9iwsa.seg.org/abstracts.
Pearson, K., 2012, Geologic models and evaluation of undiscovered conventional and continuous oil and
gas resources—Upper Cretaceous Austin Chalk, U.S. Gulf Coast: U.S. Geological Survey
Scientific Investigations Report 2012–5159, 26.
Sayers, C. M., 2010, Geophysics under stress: Geomechanical applications of seismic and borehole
acoustic waves: SEG and EAGE.
Vladimir, G., I. Tsvankin, and J. K. Cohen, 1999, Generalized Dix equation and analytic treatment of
normal-moveout velocity for anisotropic media: Geophysical Prospecting, 47, 117–148,
http://doi.org/10.1046/j.1365-2478.1999.00120.x.
© 2017 SEG Page 3996

Anisotropic prestack depth migration processing in the Permian Basin, Texas
Ananya Roy*, Lin Zheng, Xin Hu, CGG
Robin Pearson, Ernesto V. Oropeza, Sarah Mueller, Mark Chang, Anadarko Petroleum Corporation
Summary layer of salt, roughly 300 ft thick, with a top of salt (TOS)
that is rugose and poorly imaged. In addition, as in other
The Wolfcamp reservoir of the Permian Basin has recently parts of the Permian basin, the salt body contains karst
generated interest for its potentially vast oil and gas features which are formed by the dissolution of shallow
reserves. Imaging this reservoir using seismic data is evaporites by groundwater flow. Correctly understanding
challenging despite the uncomplicated geology at reservoir this high velocity and laterally complex near-surface is
depth, because of the complexity of the near-surface in this challenging. We applied an anisotropic prestack depth
region. In this work, we apply a modern anisotropic migration (APSDM) processing flow with model building
prestack depth migration (APSDM) processing flow to in depth to this Delaware Basin data. To improve on the
Permian Basin data and find that it improves the resolution legacy (2007) prestack time migration (PSTM) effort, full
of near-surface anomalies and imaging at the reservoir waveform inversion (FWI) was used to estimate the
compared to the legacy prestack time migration. overburden velocities more accurately, velocities were tied
to well markers, and processes such as 5D interpolation and
Introduction rank reduction denoise were applied to improve the S/N.
The Permian Basin lies in West Texas and South Eastern

New Mexico and it comprises the Delaware and Midland
Basins separated by the uplifted Central Basin Platform
(Figure 1). Oil and gas reserves were first discovered here
in 1921 and since then production has been steadily driven
by new discoveries and advances in extraction techniques
such as horizontal drilling and hydraulic fracturing (Texas
RRC, 2016).
Figure 2: A 2007 time migration stack from the Delaware Basin

Figure 1: A map of the Permian Basin in West Texas and South region marked by the star in Figure 1.
Eastern New Mexico (Encyclopedia Britannica, 2013). The star
marks the location of the Tunstill survey. APSDM in the Permian Basin
Figure 2 shows a time migration image from the Delaware The data presented here are from a 75 sq. mile area of the
Basin area. The Wolfcamp reservoir, which has recently Tunstill survey located in the northern section of the
generated interest for its potential (Gaswirth et al., 2016), Delaware Basin (marked by the star in Figure 1). The
lies about 12,000 ft below the surface and is on average survey was acquired in 2006 using Vibroseis. The acquired
3000 ft thick in this region. The geology at reservoir depth data have high levels of noise originating from ground roll,
seems fairly simple with little large-scale lateral variations. guided waves, and scattering from shallow evaporites. A
The near-surface, however, is quite complex. It includes a selection of data from the most challenging part of the
© 2017 SEG Page 3997

APSDM processing in the Permian Basin
survey is shown in Figure 3, where target events are Despite the challenges, the FWI-derived near-surface
completely masked by noise on the raw CDP gathers. model resolved the rugose TOS well enough to generate a
base of salt (BOS) image with very little distortion (Figure

4). The BOS is nearly flat, just as the well markers (shown
in red) suggest it should be. Since no diving rays penetrate
the large velocity contrast at the BOS, refraction FWI was
not useful below the salt body. The model below the salt
was built using reflection tomography and well log
information. A depth migration with this model tied the
well markers (Figure 4) without the need for any post-
migration processing, like a 1D stretch, to tie the wells.
Anisotropy parameters, δ and ε, were chosen to flatten
migration gathers with this velocity model. The value of δ
increases gradually from ~1% in the shallow salt to ~9% in
the Wolfcamp interval. The ratio between ε and δ is ~2.
Figure 3: Raw CDP gathers with normal moveout applied show the
high level of noise contamination in the data. Compared to the legacy PSTM flow, which compensated
the near-surface with 1D static shifts computed from
Impact of depth model building and FWI refraction tomography, APSDM using the FWI near-
surface model imaged the near-surface TOS better (Figure
Since the near-surface does not have enough useable offsets 5). FWI also improved the imaging of karsts in the salt
for curvature-based reflection tomography (Figure 3), body. These collapse-features are not always visible above
refraction tomography (Zhu et al., 2000) is typically used to the surface, and their detection using seismic data can help
estimate the near-surface velocity. In this work, refraction prevent drilling incidents. On a depth slice through the salt
FWI (Mei et al., 2014) was used as part of a depth velocity body, karst features look like isolated anomalies and can be
building flow to improve the resolution of velocity spotted on both the time and depth migration volumes
anomalies in the complex near-surface salt body. FWI was (Figure 6); however, those on the APSDM stack are clearer
challenging because the 6 Hz low frequency cut-off on the and more reliable (Figure 7). On the legacy PSTM images,
acquisition sweep made the S/N below 6 Hz poor. To avoid the BOS below the karst features is locally pulled-up,
cycle skipping with FWI in the absence of low frequencies, indicating that the karst velocity anomalies are not correctly
an initial model was built using iterative refraction resolved and the reflectors under the anomalies are less
tomography and was followed by refraction FWI from 7 Hz reliable than on the depth migration image.
to 10 Hz.
Figure 4: Depth velocity model overlaid on the 2016 APSDM stack. A zoomed-in panel shows the near-surface model from FWI, with BOS
well markers annotated as red dashes.
© 2017 SEG Page 3998

reduction denoise (Sternfels et al., 2015) and 5D

interpolation (Trad, 2007), as shown in Figure 8. The
improved noise attenuation coupled with the quadrupled

fold from 5D interpolation generated a superior image at
the Wolfcamp reservoir compared to the legacy time
migration (Figure 9). The PSDM image has better S/N, and
this can help interpretation near faults as demonstrated in
Figure 10. On the legacy time migration image, it is unclear
how far the fault extends. The fault and its vertical extent
Figure 5: TOS images for the PSTM and APSDM stacks. APSDM, are much clearer on the PSDM image.
with an FWI-derived near-surface model, images the TOS better.
Figure 8: Preliminary PSDMs before (left) and after (right) rank

reduction denoise and 5D interpolation show their impact on S/N
Discussion
Figure 6: Depth slices through the shallow salt body, 850 m above After the APSDM processing flow, CDP gathers that were
the BOS, show karst features on the 2007 PSTM stack stretched to previously dominated by noise show reflectors at reservoir
depth (top) and 2016 APSDM stack (bottom). The karsts circled in depth, but they also reveal an azimuthal jitter (Figure 11).
blue and orange are compared in Figure 7. While this azimuthal variation could be evidence of
reservoir fractures, it could also be the result of residual
errors in the overburden velocity. Even though FWI has
significantly improved the near-surface model, its accuracy
is limited by the coarse acquisition geometry of this survey.
An image of the near-surface, perpendicular to the receiver
lines (Figure 12), shows that the TOS is not imaged in the
space between the receiver lines. The coarse receiver line
spacing of 1540 ft limits our ability to clearly define the
position of the large velocity contrast at the TOS. This
limits the accuracy of our near-surface model and, in turn,
the accuracy of all the underlying reflectors. An accurate
near-surface velocity is important for decoupling the
imprint of shallow velocity errors from the properties of the
reservoir. We have shown that the accuracy of the near-
Figure 7: Cross-sections of the two karst features highlighted in surface model improves with FWI, but we also need an
Figure 6 show these anomalies are better resolved on the APSDM. acquisition geometry that is dense enough to sample the
complexities of the near-surface, at least twice the current
Impact of advanced denoise and 5D interpolation density, before we can make a meaningful analysis of more
subtle properties such as azimuthal velocity variation. Such
The 2007 legacy time processing flow included ground roll azimuthal analysis will also require a larger maximum
attenuation, several iterations of de-spike, and Radon de- crossline offset, approximately twice the existing value of
multiple for noise attenuation. In the 2016 depth processing 10,780 ft, for sufficient angle coverage at target depth.
flow, S/N was further improved by the addition of rank Finally, a low-noise broadband source (Castor et al., 2015)
© 2017 SEG Page 3999

will improve the S/N at low-frequencies which could

further improve the ability of FWI to estimate the near
surface velocity.
Figure 11: Raw gathers (shown with NMO on the left) are
dominated by noise. After depth processing, the common-offset-
common-angle (COCA) migration gather (right) shows clear
events at reservoir depth and an azimuthal jitter.
Figure 12: A near surface image, perpendicular to the receiver

lines, shows scalloping at the TOS due to coarse line spacing.
Conclusion
An APSDM workflow improved seismic imaging in the

Delaware Basin region of the Permian Basin, where the
regional geology of the Wolfcamp reservoir is simple, but
the near-surface is complex. The comparisons in this paper
show that modern denoise, interpolation, and depth model
building, including refraction and reflection tomographic
inversion and FWI, can improve resolution of the
overburden velocity and imaging at the reservoir while also
generating a realistic velocity model that ties well markers
in depth. If PSDM processing is also supported by an
acquisition geometry that is dense enough to sample the
lateral changes in the near-surface, its imprint on the
Figure 9: A comparison of the 2007 PSTM and 2016 APSDM underlying reflectors could be further reduced. This should
stacks. The APSDM includes advanced denoise processing and 5D
interpolation in addition to migration with a depth velocity model.
be the first step towards drawing better conclusions about
reservoir-scale seismic attributes in this region.
Acknowledgments
We would like to thank our colleagues, Erasmo Rodriguez,

Tunde Markos, Jiawei Mei, Jeshurun Hembd, Yan Huang,
and Sue Rezai at CGG for help with processing and for
valuable discussions. We would also like to thank our
colleagues at Anadarko, Stella Braz and Alan Gunnell, for
their feedback on processing and Tim Fasnacht for valuable
suggestions on this paper. We are grateful to the CGG
Multi Client Data Library for providing us use of the data
Figure 10: Legacy PSTM (left) and APSDM (right) migration and the Legacy PSTM volume.
images near a fault just below the reservoir interval.
© 2017 SEG Page 4000

EDITED REFERENCES
REFERENCES
Castor, K., T. Bianchi, O. Winter, and T. Klein, 2015, Efficient harmonic-distortion mitigation on vibro-
seismic sources: Presented at the EAGE Workshop on Broadband Seismic, Extended Abstracts,
BS08.
Encyclopedia Britannica, 2013, Permian Basin, https://www.britannica.com/place/Permian-Basin,
accessed 1 March 2017.
Gaswirth, S. B., K. R. Marra, P. G. Lillis, T. J. Mercier, H. M. Leathers-Miller, C. J. Schenk, T. R. Klett,
P. A. Le, M. E. Tennyson, S. J. Hawkins, M. E. Brownfield, J. K. Pitman, and T. M. Finn, 2016,
Assessment of undiscovered continuous oil resources in the Wolfcamp shale of the Midland
Basin, Permian Basin Province, Texas, 2016: U.S. Geological Survey Fact Sheet 2016-3092.
Mei, J., S. Ahmed, A. Searle, and C. Ting, 2014, Application of full waveform inversion on Alaska land
3D survey: 84th Annual International Meeting, SEG, Expanded Abstracts, 981–986,
http://dx.doi.org/10.1190/segam2014-0505.1.
Sternfels, R., G. Viguier, R. Gondoin, and D. Le Meur, 2015, Joint low-rank and sparse inversion for
multidimensional simultaneous random/erratic noise attenuation and interpolation: 77th Annual
International Conference and Exhibition, EAGE, Extended Abstracts, We N112 04.
Texas RRC., 201, Permian Basin Information, http://www.rrc.state.tx.us/oil-gas/major-oil-gas-
formations/permian-basin/, accessed 1 March 2017.
Trad, D., 2007, A strategy for wide-azimuth land interpolation: 77th Annual International Meeting, SEG,
Expanded Abstracts, 946–950, http://dx.doi.org/10.1190/1.2792562.
Zhu, T., S. Cheadle, A. Petrella, and S. Gray, 2000, First-arrival tomography: Method and application:
70th Annual International Meeting, SEG, Expanded Abstracts, 2028–2031,
http://dx.doi.org/10.1190/1.1815839.
© 2017 SEG Page 4001

Geophysics and its Role in Appalachian Basin Oil and Gas Exploration and Development
Joel Starr*, EQT Production
Summary
Trenton-Black River Play and renewed interest in
The Appalachian Basin, as an oil and gas province, runs Seismic
from southern New York State to Tennessee. Since the
1970s, geophysics has played an important role in allowing After the collapse of oil prices in the mid 1980’s the
for enhanced exploitation of the oil and gas resources in this acquisition of seismic data dropped dramatically. Most of
region. Initial efforts were limited to mostly gravity and 2D the major oil companies abandoned the basin and traded
seismic but recently, with focus on the shales, horizontal their databases to seismic data licensing companies. The
drilling and hydraulic fracturing, 3D seismic along with basin was left to smaller operators who continued to
micro-seismic and distributed acoustic sensing (DAS) are explore using the previously acquired seismic data. While
playing an increasingly important role in improving some prospect lines and a few key-hole 3Ds were acquired
hydrocarbon extraction. This presentation reviews the during this time, the massive investments of the ‘70s and
relationship of geophysics to the boom and bust cycle of early ‘80s were over.
hydrocarbon development in the Appalachian Basin over the
last 47 years. In the late 1990s success with wells drilled in the Trenton-
Black River began a rebound in seismic activity. One of
Modern 2D seismic and the Cambro-Ordovician Play the best documented Trenton-Back River fields is Saybrook
field discussed by Sagan and Hart 2006. The play concept
Ever since the success of “Drake’s Well” in 1859, oil and was to identify sag features on seismic data which would be
gas exploration in the Appalachian Basin has been indicative of strike-slip grabens where hydrothermal waters
punctuated with periods of extremely intense activity had dolotomized the Trenton limestone creating secondary
followed by relatively negligible interest (Yergin 1991). porosity. As oil prices recovered in the early 2000s activity
These boom and bust cycles have largely been due to market in the Trenton-Black River continued to grow.
forces but have also been influenced by technology. The
Appalachian Basin (Figure 1) was the center of oil and gas Marcellus Shale and large scale 3D Acquisition
exploration in the United States until Spindletop was
discovered in 1901, after which the focus of drilling activity It is said success begets success, this certainly could be
gradually shifted to the Gulf of Mexico. Interest in the attributed to the Marcellus Shale and the massive effort put
Appalachian Basin waned as the use of existing technology into the acquisition of 3D seismic in Pennsylvania and
matured. West Virginia since 2008.
Prior to 1970, it was well understood that only a relatively With the continued success of the dolotimized Trenton-
shallow portion of the Appalachian Basin had been tested. Black River play, many operators turned their attention to
The challenge for explorationist of the time was to prospect fractured Trenton for porosity and permeability. Range
for deeper reserves without incurring the large cost of Resources was targeting fractured Trenton in Washington
drilling deep exploration wells. In 1970, 2D seismic had County PA when they had a large show in the Marcellus
evolved to the point where deeper structures of the Cambrian Shale. Drawing off the success of the Barnett Shale in
to Ordovician could be evaluated. A combination of Texas, Range completed the Marcellus section in 2004
vibroseis, crooked line geometry, digital processing and fold which was the beginning the current efforts to target
had matured to the point that a large scale seismic effort Marcellus Shale.
could be made to explore the deeper sections of the basin.
From 1971 to 1973 over 12,000 linear miles of 2D seismic As other operators raced to duplicate Range’s success it
were acquired. In 1973 the Arab oil embargo began and oil became evident that not all shale is the same. Geohazards
prices rose (Figure 2) until their collapse in 1984. During such as faults and extreme changes in dip had to be
this time an additional 20,900 miles of 2D were acquired in anticipated. Large scale multi-client 3D programs proved
the basin. Figure 3 is an example of a 2D seismic line an efficient and cost effective solution. As of 2016 over
showing Cambro-Ordovician section relative to shallower 7000 square miles of multi-client 3D seismic has been
major play targets. acquired in Pennsylvania, West Virginia and Ohio.
Acquiring 3D seismic in the Appalachian Basin presents its

own special challenges. The tracts tend to be small and
© 2017 SEG Page 4002

governance is divided between the state, county, townships Distributed acoustic sensing (DAS) shows great promise in
and city or borough. Properly resourcing the permitting improving our understanding of how these wells produce. A
phase of any seismic program is critical to the success of a fiber optic cable is cemented into the casing of the well and
project. Most 3D seismic programs tend to be a mix of can be used to monitor the stimulation and production of the
truck mounted vibroseis, along the passable roads, and shot well. In addition to monitoring the well that contains the
holes in the open fields. Cabled recording systems do not fiber optic cable, DAS can be used as a cross-bore hole strain
lend themselves to the area due to the topography and the monitor of adjacent wells as they are stimulated.
fractious nature of permitting. Modern autonomous
recording systems such as single group recorders have Interpreting microseismic and DAS data as standalone
proven to be far more effective than cabled systems. datasets can be challenging. They are often ambiguous by
themselves. 3D seismic adds context to the interpreted
It is a rare event for a receiver or source location to be results and can enhance the value of individual datasets.
utilized at is its pre-plotted location, most surface points are Well defined and repeatable trends in microseismic data are
skidded to a location that the permitting and topography often associated with regions where the geology is unvarying
allow. As a result, processing technology that can along the well bore. Erratic results can often be attributed to
regularize the acquisition geometry such as 5D structural complexities such as faults and fractures. Having
interpolation is heavily utilized. independent datasets which contribute to a single
interpretation improves the confidence in the engineering
Microseismic and Distributed Acoustic Sensing recommendations that are made from the data.
Geophysics covers a broad spectrum of technologies used to Conclusions

enhance oil and gas exploitation. It is a means to an ends,
not the ends. It is important to understand that when Geophysics has played an important role in the continued
evaluating the new problems associated with new play development of oil and gas resources in the Appalachian
concepts. Depth imaging technology in the Gulf of Mexico Basin. Developments in seismic technologies of the early
was primarily driven by the need for reliable sub-salt 1970s allowed for exploration of the deeper plays. As these
imaging. Shale production in the Appalachian Basin is plays required large capital investments, the level of activity
dependent on targeting the most productive units and was closely tied to the commodity price for oil and gas.
understanding how the shales will respond to hydraulic
stimulation. Hence accurate structural images and rock Each play had unique challenges, the geophysics had to
mechanics are the focus of geophysical efforts in adapt to those changes. The Trenton-Black River play is
Appalachian Shale development. strongly dependent on identifying sags associated with
graben structures. Shale development is more dependent on
Microseismic monitoring has emerged as a key technology identifying geohazards for drilling horizontal laterals and
in determining where, when and how the formations are determining the mechanical properties of the target
responding to hydraulic stimulation. Being a relatively new formation for proper calibration of hydraulic stimulation.
technology, there has been rapid improvement in the use of
this method. Originally used to identify the location and Acknowledgements
timing of rock failure events, techniques such moment tensor
inversion can determine the slip vectors for each event. The author would thank SEI, TGS, Geokinetics, Seitel and
From these slip vectors the relative magnitude and CGG for providing statistics about seismic data acquired in
orientation of the in-situ stress fields can be estimated. the Appalachian Basin. He would also like to thank IHS
Statistical analysis such as diffusivity are used to determine for providing the oil price history.
the progress of the water front as it diffuses through the rock.
© 2017 SEG Page 4003

Figure 1: Regional map of the Appalachian Basin showing available 2D seismic as light orange lines and 3D seismic surveys
are outlined with dark green lines.
Figure 2: 2D and 3D seismic acquisition activity relative to the price of oil by year. Major play focus for a given time period
is highlighted with colored boxes.
© 2017 SEG Page 4004

Figure 3: Seismic stratigraphy of the Cambro-Ordovician relative to shallower target sections in the Appalachian Basin.
© 2017 SEG Page 4005

EDITED REFERENCES
REFERENCES
Sagan, J. A., and B. S. Hart, 2006, Three-dimensional seismic-based definition of fault-related porosity
development: Trenton-Black River interval, Saybrook, Ohio: AAPG Bulletin, 90, 1763–1785,
http://dx.doi.org/10.1306/07190605027.
Yergin, D., 1991, The Prize: The Epic Quest for Oil, Money and Power: Simon and Shuster, 875.
© 2017 SEG Page 4006

Seismic reservoir characterization of Utica-Point Pleasant shale with efforts at quantitative
interpretation– a case study
Satinder Chopra*†, Ritesh Kumar Sharma†, Hossein Nemati† and James Keay+
†Arcis Seismic Solutions, TGS, Calgary; +TGS, Houston
Summary available well data and understand the parameters that populate
Utica shale is one of the major source rocks in Ohio and extends the reservoir intervals at the location of the wells. The sonic,
across much of eastern US. Its organic richness, high content of density, gamma ray, resistivity, porosity well log curves are
calcite, and development of extensive organic porosity makes it a sought for the available wells over the 3D seismic volume. Core
perfect unconventional play and has gained the attention of the oil analysis results, geochemical as well as geomechanical data are
and gas industry. The primary target zone in the Utica includes available for one well.
Utica, Point Pleasant, and Trenton intervals. In the present study,
3D seismic data acquisition and processing
we attempt to identify the sweet-spots within the Point-Pleasant
interval using 3D seismic data, available well data, and other The acquisition of a 702 mi2 (1818 km2) 3D seismic survey spread
relevant data. This has been done by way of organic richness and over Carroll, Tuscarawas, Guernsey, Noble, Belmont, Harrison
brittleness estimation in the rock intervals. The organic richness and Jefferson counties of eastern Ohio, was completed in late
is determined through TOC content which is derived by 2015. The survey falls in the wet gas and light oil windows of the
transforming the inverted density volume. The core-log Utica-Point Pleasant. The acquisition parameters include 220 ft
petrophysical modeling provides the necessary relationship for (67.056 m) for source and receiver intervals, 660 ft (201.168 m)
doing so. The brittleness is derived using rock-physics parameters for receiver line spacing, 1320 ft (402.336 m) source line spacing,
such as Young’s modulus and Poisson’s ratio. Deterministic maximum offset as 19,186 ft (5847.89 m), 2 ms sample interval,
simultaneous inversion along with a neural network approach are 5 s record length, which yielded a bin size of 110 ft by 110 ft (33.5
followed in order to compute rock-physics parameters and density m X 33.5 m). Two vibrator sweeps of 16 s are used as the seismic
using seismic data. The consistency of sweet spots identified source. The processing of this large data volume was completed
based on the seismic data with the available production data in June 2016, with anisotropic prestack time migration (PSTM)
emphasize the integration of seismic data with all other relevant gathers and stacked volume with 5D interpolation made available
data. for reservoir characterization and quantitative interpretation.
Introduction Well-log correlation
The Utica shale is considered a source rock for oil and natural gas, Correlation of well log information with 3D surface seismic data
which migrated upwards and were produced by conventional is a convenient way to extend the measured rock properties at well
means in the overlying rock formations. According to a 2012 locations spatially over the 3D volume.
USGS report, the formation holds 940 million barrels of oil and
As we started collating well data for our study we realized that
approximately 38 tcf of natural gas (Kirschbaum et al., 2012), but
the wells that had density curves are located in a cluster to the
with more drilling and production, these estimates have been
northern part of the survey, and very few wells had both sonic and
revised and stand at 2 billion barrels of oil and 782 tcf of natural
density curves. A frequently encountered situation is when not
gas (Cocklin, 2015). The thermal maturity studies in the Utica
many wells have shear sonic log curves available. It is always
shale have indicated a northeast to southwest trend over eastern
desirable to have a uniform location of wells with sonic, density
Ohio and western Pennsylvania, with a western oil phase window,
and other curves (GR, porosity, resistivity, etc.) though sparse, as
a central wet gas phase window and an eastern dry gas phase
it helps with the generation of a reliable low-frequency
window.
impedance model for impedance inversion, as well as for carrying
We present our attempts at seismic reservoir characterization of out any neural network analysis for computation of a reservoir
the Utica – Point Pleasant package in eastern Ohio. Beginning property. Besides, any crossplotting carried out on well data
with a discussion about the data that are available for the exercise, located sparsely on a 3D volume, and in localized clusters may
next we describe the workflow that is followed. The goal of not be a true representation of relationships between the
seismic reservoir characterization is essentially the identification crossplotted variables. We therefore selected wells that had an
of sweet spots that represent the most favorable drilling areas. optimum distribution as shown in Figure 1. Some of the wells
Such an exercise entails understanding the elastic properties of located at the edge of the 3D survey were projected a little bit
the reservoir intervals, lithology, fluid content and their areal inside, as the seismic data close to the edges of the survey are not
distribution. A good starting point for doing this is to use the very trustworthy.
© 2017 SEG Page 4007

Seismic reservoir characterization of the Utica-Point Pleasant shale
Once the final seismic data are loaded on the workstation, we The organic richness in the shale rocks influences properties such
assessed its quality and frequency content. The data were as compressional and shear velocities, and density. Therefore,
preconditioned for random noise attenuation by putting it through attempts have been made to detect changes in TOC from the
structure-oriented filtering (Chopra and Marfurt, 2007; Marfurt, surface seismic response using impedance and other attributes
2006). such as VP-VS ratio, Lambda-rho, Mu-rho etc. (Sharma and
Chopra, 2016). In this study we have tried to bring in data from

In Figure 2 we show the correlation of the sonic, density and GR
core analysis, as well as geochemical and geomechanical
log curves and synthetic seismograms for well W-3 with the
analysis, and integrate that with surface seismic data. The density
seismic data. Five horizons corresponding to Trenton Limestone,
and TOC measurements made on the core samples in the Point
Point Pleasant, and Above-Utica in our zone of interest, and
Pleasant interval were crossplotted as shown in Figure 5. A
Clinton sandstone and Onondaga limestone above it were picked
strong linear relationship is seen between them. This suggests
and are indicated in Figure 2. While the Trenton and Onondaga
that the density attribute would be required if the organic-rich
limestones show good contrast at their levels on the log curves
zones in the Point Pleasant interval are to be determined from
(and thus prominent reflections on seismic), reflections
seismic data.
corresponding to Point Pleasant and Clinton sandstone were also
pickable. But no individual reflection corresponding to Utica And as stated above, as the angle range was not favorable for
shale could be picked, and so the closest pickable reflection was computing density from seismic data through simultaneous
considered and called ‘Above-Utica’. A zero-phase wavelet was inversion, we turned to neural network analysis for its
extracted from the seismic data using a statistical process (shown determination. We decided to determine density with
on the top) and was used for generating the synthetic seismogram. probabilistic neural network analysis, employing amongst others
An overall good correlation is seen between the two. A some of the attributes determined from simultaneous inversion.
representative seismic section from the 3D seismic volume The details of the neural network approach followed, as well as
passing through two wells is shown in Figure 3. Good correlation some other work with regard to the integration of core,
is seen between the impedance curves and the seismic. geochemical and geomechanical data and seismic data are being
presented in a companion paper (Sharma et al., 2017). But in
As shear log curves were available in only three of the eight wells
Figure 6 we show how the predicted density compares with the
over the 3D survey, that had sonic and density curves, we
measured density at the location of well W-7. The good match
crossplot the P- and S-impedance for these three wells as shown
between the curves enhanced our confidence in this approach.
in Figure 4. The linear relationship seen therein is used to
generate the shear curves for five wells that didn’t have the shear Once the density volume was determined from neural network
curves, and then used those seven shear impedance curves (3 analysis, the linear relationship shown in Figure 5 was used to
measured and 4 predicted using the relationship) for the transform it to a TOC volume. High TOC content was noticed in
generation of the low-frequency S-impedance model for the northern part of the survey, which is consistent with TOC
inversion. Using the low-frequency model generated with a single trend observed in the Utica-Point Pleasant play (Wickstrom,
well as one of the inputs, and some other seismic data volumes 2013).
(relative acoustic impedance, instantaneous amplitude, dominant
frequency, filtered seismic (10-20-30-40 Hz and 10-20-50-60 Besides the organic richness consideration, it is vital that reservoir
Hz), a multiregression approach (Ray and Chopra 2015, 2016) is zones are sufficiently brittle as fracturing potential of a shale
used, wherein a target log is modeled as a linear combination of reservoir is a fundamental function of its brittleness. Attempts are
several input attributes at each sample point. This approach usually made to identify the brittle zones with the help of
results in more accurate low-frequency impedance models. Poisson’s ratio and Young’s modulus as a rock’s ability to fail
Having determined the low-frequency models for both P- and S- under stress is represented by the former, while the ability of
impedance, the next step was to carry out preconditioning of the sustaining fractures is reflected by the latter.
prestack data for enhancing its signal-to-noise ratio. We found
that the useable angle range was 34o, and thus the output attributes We crossplot Young’s modulus and Poisson’s ratio for data for
were P-impedance and S-impedance, but the density attribute wells W-1 and W-7 from the Utica through Point Pleasant to the
could not be determined with simultaneous inversion. Trenton interval. We notice a positive correlation between the
two parameters and noticed that the Point Pleasant interval
Sweet spot determination exhibits low Young’s modulus and Poisson’s ratio relative to the
Utica interval. As is seen in many other shale formations,
The main goal for shale resource characterization is usually the
brittleness is found to increase as Poisson’s ratio decreases and
identification of sweet spots which represent the most favorable
Young’s modulus increases (Rickman et al., 2008). Point
drilling targets. Such sweet spots can be picked up as those
Pleasant does not seem to follow this behavior, even though the
pockets in the target formation that exhibit high total organic
production from the multistage fracking in this interval has been
carbon (TOC) content, high porosity, as well as high brittleness.
established (Patchen and Carter, 2015). Within the Point Pleasant
© 2017 SEG Page 4008

interval we see a variation in these two parameters. To study the and geomechanical data. This has been done by deriving rock-
variation of these parameters within this interval, we restrict the physics parameters (Young’s modulus and Poisson’s ratio)
data points coming into the crossplot to just the Point Pleasant through deterministic simultaneous inversion and neural network
interval and see that the cluster of points coming from W-7 (to the analysis. We find that the Point Pleasant formation does not seem
south) and the ones coming from W-1 (to the north), Poisson’s to follow the commonly followed variation in terms of low
ratio and Young’s modulus both decrease going from north to Poisson’s ratio and high Young’s modulus for brittle pockets.
south. Instead, by restricting the values of Poisson’s ratio and examining
the variation of Young’s modulus, we are able to determine the
Grieser and Bray (2007) proposed computing a brittleness brittleness behavior within the Point Pleasant interval. Combining
the brittleness behavior with the organic richness determined
average from Young’s modulus and Poisson’s ratio and
through the TOC content, we are able to pick sweet spots in the
demonstrated deciphering brittle and ductile shale pockets within
Point Pleasant interval which match the production data.
the Barnett shale by considering all values of Poisson’s ratio less
than 0.25 as threshold and all values of Young’s modulus greater
Through this case study, we emphasize the integration of 3D
than 3.1 x 106 psi.
surface seismic data with all other relevant data so as to accurately
characterize the Point Pleasant formation.
We follow a similar approach and demonstrate its application to
the Utica-Point Pleasant play. Realizing that the Point Pleasant
Acknowledgements
interval has higher calcite content, and therefore its ability to fail
under stress and sustain fractures must be high, we picked up the We wish to thank Arcis Seismic Solutions/TGS for encouraging
P-impedance and S-impedance derived from simultaneous this work and also for the permission to present and publish it.
inversion and density derived from probabilistic neural network
analysis to compute Young’s modulus and Poisson’s ratio
attributes. These are then crossplotted just for the Utica to
Trenton interval as shown in Figure 7. Notice all points below
the value 0.23 for Poisson’s ratio (enclosed in red and green
polygons) come from the Point Pleasant interval. Thus we Figure 1: Picked
horizon at the Point
interpret this interval to be prone to get fractured under stress.
Pleasant level
The ability of this interval to sustain fractures in a relative sense indicating the
can be examined based on Young’s modulus attribute. It can be dipping reflections
seen from Figure 7a that the points enclosed by the green polygon from northeast to
southwest. The
correspond to higher values of Young’s modulus, than the points locations of the
enclosed by the red polygon. When we project these points on the available wells 1 to
vertical arbitrary line passing through the wells, as exhibited in 7 are also indicated.
Figure 7b, we notice that the northern side of this line exhibits
higher brittleness than the southern side.
To examine the lateral variation in the Young’s modulus we draw

a horizon slice from the Young’s modulus volume which is shown
in Figure 8a. The northern part of the display shows higher values
of Young’s modulus. Thus by restricting the values of Poisson’s
ratio and examining the variation of Young’s modulus, we have
been able to determine the variation in the brittleness of the Point
Pleasant interval. In addition to brittleness, organic richness was
also examined through the TOC volume. For doing this we draw
an equivalent horizon slice from the TOC volume which is shown
in Figure 8b. We notice higher TOC values in the northern part.
The areas highlighted in the black polygons are thus the sweet
spots that have been determined from the above analysis. This
seems accurate enough as confirmed by the available production
data overlaid on the TOC display.
Conclusions
Figure 4: Crossplot of P-impedance versus S-impedance using
We have characterized the Point Pleasant formation in eastern well-log data from three wells 1, 4 and 7. A high correlation
Ohio using 3D surface and its integration with core, geochemical coefficient is seen for the linear trend observed.
© 2017 SEG Page 4009

Figure 2: Correlation of well W-3 P-velocity, density and

Gamma ray curves with seismic data. Notice the sharp
impedance contrast seen at the Onandaga Limestone and
Rochester Shale levels giving rise to strong reflections. The Figure 3: A representative inline running east-west from the 3D seismic volume with P-velocity
horizons corresponding to Onandaga Limestone, ‘Clinton’ curves overlaid on it The quality of the seismic data is good. The horizons picked on the line show
sandstone and Point Pleasant and Trenton levels are pickable and the Onandaga Limestone, ‘Clinton’ sandstone, Above-Utica, Point Pleasant and the Trenton to the
seen clearly on the seismic. (Data courtesy: TGS, Houston) right. (Data courtesy: TGS, Houston)
Figure 7: (a) Crossplot

between Young’s modulus and
Poisson’s ratio as derived from
seismic data for the Utica to
Trenton interval. (b) The
cluster points in different
polygons on the crossplot
when projected on the vertical
show higher brittleness on the
northern side of the survey.
(Data courtesy: TGS, Houston)
Figure 5: Crossplot of density and TOC as determined

from core data in the Point Pleasant interval. A good
linear relationship is seen between the crossplotted
variables.
Figure 8: Horizon
slices from (a)
Young’s modulus,
and (b) TOC
volumes, both
averaged in a 10 ms
window in the Point
Pleasant interval. The
highlighted portions
indicate the sweet
spots corresponding
to high Young’s
modulus and high
TOC. Overlaid on the
Figure 6: The density trace predicted with
TOC display is the
neural network application compared with the
production data.
measured density log curve at the location of
(Data courtesy: TGS,
well W-7. The two curves overlay well and
Houston)
thus enhance our confidence in neural network
© 2017 SEG
density prediction. Page 4010
EDITED REFERENCES
REFERENCES
Chopra, S., and K. J. Marfurt, 2007, Seismic attributes for prospect identification and reservoir
characterization, Geophysical Development Series, SEG,
https://doi.org/10.1190/1.9781560801900.
Grieser, B., and J. Bray, 2007, Identification of production in unconventional reservoirs: SPE 106623
(SPE Production and Operations. Symposium), Oklahoma City, OK, March 31-April 3, 2007.
Cocklin, J., 2015: Shale Daily, http://www.naturalgasintel.com/articles/102982-wvu-study-finds-bounty-
of-utica-shale-natgas-waiting-for-production (accessed on February 18, 2017).
Kirschbaum, M.A., C. J. Schenk, T. A. Cook, R. T. Ryder, R. R. Charpentier, T. R. Klett, S. B. Gaswirth,
M. E. Tennyson, and K. J. Whidden, 2012, Assessment of undiscovered oil and gas resources of
the Ordovician Utica Shale of the Appalachian Basin Province, 2012: U.S. Geological Survey
Fact Sheet 2012–3116, 6.
Marfurt, K. J., 2006, Robust estimates of reflector dip and azimuth: Geophysics, 71, P29–P40.
Patchen, D.G. and K. M. Carter, eds., 2015, A geologic play book for Utica Shale Appalachian basin
exploration: Final report of the Utica Shale Appalachian basin exploration consortium, 187,
http://www.wvgs.wvnet.edu/utica (accessed on 18th February, 2017).
Ray, A. K. and S. Chopra, 2015, More robust methods of low-frequency model building for seismic
impedance inversion: 85th Annual International Meeting, SEG, Expanded Abstracts, 3398–3402,
https://doi.org/10.1190/segam2015-5851713.1.
Ray, A. K. and S. Chopra, 2016, Building more robust low-frequency models for seismic impedance
inversion: First Break, 34, 29–34, https://doi.org/10.3997/1365-2397.2016005.
Rickman, R., M. J. Mullen, J. E. Petre, W. V. Grieser, and D. Kundert, 2008, A Practical Use of Shale
Petrophysics for Stimulation Design Optimization: All Shale Plays Are Not Clones of the Barnett
Shale. SPE 115258, Society of Petroleum Engineers.
Sharma, R. K., and S. Chopra, 2016, Identification of sweet spots in shale reservoir formations: First
Break, 34, 39–47, https://doi.org/10.3997/1365-2397.2016012.
Wickstrom, L., 2013, Geology and activity of the Utica-Point Pleasant of Ohio,
http://www.searchanddiscovery.com/documents/2013/10490wickstrom/ndx_wickstrom.pdf
(accessed on 18th February, 2017).
© 2017 SEG Page 4011

Unlocking the secrets of North American shale reservoirs using deterministic rock physics
models
Rob Holt, Occidental Oil and Gas Corporation
Summary porosity sediments as here fluid effects often have a greater

impact on the elastic rock properties than they do in tighter
I present a new workflow that has been used to build rocks, as shown for example by Ødegaard & Avseth
detailed data-driven rock physics models of various (2003). In contrast, Davie et al. (2012) showed that simple
prospective unconventional shale intervals within the empirical models may be used to predict facies from
Permian Basin of the US, and the Horn River and Western seismic AVO inversion data at a regional scale for
Canadian Sedimentary Basins of Canada. A couple of unconventional shale reservoirs in the Horn River Basin of
simple approximations enable us to estimate depth-varying Canada. It may be inferred from this study that the
in-situ geophysical properties (Vp, Vs and density) for the variations in the observed elastic rock property of these
three commonly-defined and often volumetrically- shales were primarily driven by changes in mineralogy (aka
dominant mineral groups – carbonates, clays and silicates. facies) and that the effects of microstructural (porosity and
Confidence that these estimated mineral properties are pore shape) variations were minor in comparison. A
physically reasonable is obtained by: (1) testing using similar finding from an unrelated study was published by
synthetic data; and (2) comparison with published data. Saxena et al. (2016).
After the detailed rock physics models have been built, we With any model, it is critical to understand the
can test whether simplified models may be usefully approximations upon which it is based, as well as the
applied, to either geophysical well log data or seismic AVO uncertainties of both the input data and the output
inversion data, to predict mineralogy for regional reservoir predictions. A pragmatic use of rock physics models to
characterization studies. predict facies is achieved using Bayesian reservoir
characterization technique (e.g. Ball, V., et al., 2016). In
Rock physics model building for reservoir the Bayesian approach, input data uncertainties (defined
characterization using the probability distribution functions of each facies)
are used as a priori inputs for the facies prediction.
At the microscopic scale, complex theoretical rock physics Incorporating uncertainties directly into the Bayesian
models are required to model the elastic rock properties of reservoir characterization workflow enables probability
different lithofacies. Subsurface factors that need to be maps to be output for each of the predicted facies.
taken into consideration at this scale include the rock
mineral frame, microstructure (porosity and pore shape), Alternatively, deterministic approaches for predicting
pore fill, temperature and pressure. Practical application of reservoir properties are based on the principal that a set of
theoretical models requires simplifying approximations. observed rock properties may be modeled as a function of
the reservoir properties:
At the macroscopic level, relationships between facies and
elastic rock properties may be derived empirically using observed rock properties = function (reservoir properties)
well log data, then applied to predict facies from elastic
rock properties obtained from AVO inversion of 2D or 3D and that the reservoir properties may therefore be estimated
seismic data. As with theoretical models, simple empirical from the observed rock properties using the inverse
models are only applicable under a limited set of function:
circumstances, not least when they are built to fit just two
estimated reservoir properties = function-1(rock properties)
measured elastic rock properties, P-Impedance and Vp/Vs
(or combinations thereof such as Lambda-rho, Mu-rho, Geophysical applications of inverse functions (constructed
Poisson’s Ratio, S-Impedance, etc.). using inversion) are widespread, and their success depends
upon how well formulated they are. A classical application
An early example of applying macroscopic empirical rock in the field of reservoir characterization, with many years
physics models for predicting facies was presented by of successful use, is that of Thomas & Stieber (1975).
Goodway et al. (1997). In this study, the authors showed Here, a simple linear model as shown in Figure 1 is used to
how the higher porosity target gas-charged conventional predict key reservoir properties - the ratio of sand vs. shale
reservoir sands separated from lower porosity sands in and porosity, from measurements of gamma ray and
elastic rock property space, and they exploited this neutron porosity.
separation for a successful drilling campaign. Models can
get significantly more complex in less consolidated, higher
© 2017 SEG Page 4012

Unlocking the secrets of North American shale reservoirs
Figure 1: Application of Thomas & Stieber (1975) for estimating

the sand vs. shale ratio and porosity from gamma ray and neutron Figure 2: sCore facies classification scheme plotted on the classical
porosity log measurements; from van der Wal & Stromberg ternary diagram of silicates, clays and carbonate minerals, from
(2012). Gamero-Diaz et al. (2012).
Criticism of the deterministic inversion framework is based If we can assume that a close relationship between facies
on two key arguments: (1) that “it assumes that the inputs and mineralogy such as this generally holds for
are known precisely and are not subject to errors”, and (2) unconventional shale reservoirs, it is logical to ask whether
that it fails to “take into account the uncertainty in the we can predict the relative fractions of the three dominant
match between the observed seismic data and the forward mineral groups (instead of facies) from measurements of
model” (Ball, V., et al., 2016). In defense of their elastic rock properties. This would be useful, for example,
workflow, Thomas & Stieber (1975) countered these when empirical relationships are observed between
arguments by stating that: (1) “Errors have been mineralogy and reservoir properties of economic interest
introduced into the model by many of our assumptions. such as porosity or water saturation.
However these assumptions are necessary to arrive at the
simple equations presented”; and (2) “The small Holt & Westwood (2016) demonstrated that simple
uncertainties introduced ... as we build our model are deterministic models may be successfully constructed using
within the uncertainties in typical field applications and are inversion to predict mineral fractions from elastic rock
therefore justified to simplify the treatment of the data”. In properties (and vice-versa) at the well log scale, as shown
other words, they believed that the simplicity of their in Figure 3a, for a couple of North American shale
approach, and the value derived from their results, trumped reservoirs. Their workflow has similarities to the
the concerns about the inherent errors and limitations of the workflows of Alvarez et al. (2015), Davie et al. (2012), and
deterministic approach. Thomas & Stieber (1975), and is fundamentally based on
the volumetric-weighted averaging equation:
A simple deterministic way to predict shale reservoir
mineralogy from elastic rock properties Measured property = propertycarbonate.fractioncarbonate
+ propertyclay.fractionclay
+ propertyquartz.fractionquartz
With the publication of the sCore facies classification
scheme, Figure 2, Schlumberger formalized a close Here the measured property is defined to be the fraction-
relationship between shale (aka mudstone) facies and the weighted average of the effective properties of the three
three main mineral groups – here called carbonates, clays dominant mineral groups.
and QFM. In this classification scheme, for example,
facies described are being carbonaceous are defined as Holt & Westwood (2016) also demonstrated that simple
being largely composed of carbonate minerals (Calcite and models, calibrated at the log scale, could be applied to pre-
Dolomite); facies described as siliceous are defined as stack seismic AVO inversion data, effectively transforming
being largely composed of silicate minerals (Quartz, the seismic data to geological sections, as shown in Figure
Feldspar and Mica); and facies described as argillaceous 3b. This makes seismic data significantly more accessible
are defined as being largely composed of the various clay to a wide audience of subsurface and managerial
minerals. Minor minerals are ignored. professionals.
© 2017 SEG Page 4013

Figure 3: [A] Log scale mineral fractions predicted from elastic rock properties (red) compared to minerals determined from Litho Scanner logs
(blue). Here a simple linear model gives good overall predictions over a significant depth range (hundreds of meters). [B] Seismic-scale quartz
fraction predicted from pre-stack AVO inversion data. Potential higher porosity reservoir units have a higher quartz content (shown in yellow
and green) than other stratigraphic levels - see for example Munson (2015). Modified from Holt & Westwood (2016).
A number of valid concerns have prevented widespread ii. The information derived from application of the
application of such simple deterministic workflows: workflow needed to be actionable – either by yielding
unique subsurface insight and/or generating models
 First, it is challenging to demonstrate that the model that can be used at the well log and seismic scales.
properties (e.g. the estimated effective AI and Vp/Vs
values for the three mineral groups) are physically As in the original simple workflow, the model is built using
meaningful. It is certainly possible to generate a well log calibration data – P- and S-wave sonic, density and
number of different models that give apparently good mineral logs. Two key differences are:
predictions of mineralogy from measured elastic rock
properties. If a model that is not physically reasonable 1. Instead of inverting for two effective properties (e.g.
is constructed during inversion, then, although the AI and Vp/Vs) for each of the three dominant mineral
results may look good at the calibration locations, they groups, pairs of effective properties (Vp and Vs; Vp
are likely not to be elsewhere. and density; and Vs and density) are now estimated
 Secondly, whilst it is quite possible to build simple simultaneously for each of the three dominant mineral
models to predict mineralogy from elastic rock groups. Testing with synthetic data showed that this,
properties, questions naturally arise as to where, when at the very least, yields accurate mineral property
and how the underlying simplification of the
estimates for the two most common mineral groups.
subsurface is justified (the elastic rock properties of
2. The predicted effective mineral properties are allowed
the subsurface are not just driven by the three
dominant mineral groups, but also by changes in to vary smoothly and continuously with depth.
porosity, pore fluids, effective pressure, the presence Plotting variations in these properties as a function of
of other minerals, and other factors). depth reveals the richness of information available to
be extracted about the subsurface and yields new
A new, robust deterministic inversion for in-situ insight into variations in reservoir properties. These
mineral property estimation are perhaps the first estimates of in-situ mineral
velocities and densities?
Given the success of the simple deterministic workflow of
Holt & Westwood (2016), whilst being mindful of its Note that this inversion is run unconstrained, with no
limitations and of the valid concerns outlined previously, a conditions imposed on the velocities or densities estimated
new, more robust deterministic workflow has been for the three mineral groups. To avoid errors when there
developed. During the development process, two key are more than three significant mineral groups present, data
criteria were fulfilled: are excluded from the inversion when the volumetric sum
of the three dominant mineral groups is less than, say, 95%.
i. The estimated effective mineral properties needed to To avoid noise contamination, outliers may be clipped or
be demonstrably physically reasonable. excluded from the analysis.
© 2017 SEG Page 4014

Figure 4 shows the application of the new workflow to data predictions are similar to published North Sea data,
from the Wolfcamp interval of a well from the Permian and the carbonate prediction trends are similar to
Basin. At the top I show a comparison of mineral fractions separately published calcite data.
predicted from the model (using Vp, Vs and density logs as iv) The AI vs. SI trends for quartz and carbonate are
the prediction input data) and the mineral fractions consistent with the sand and carbonate trends
determined from an ECS log. In the middle, I show a published for the Western Canadian Sedimentary
comparison of the well log sonic and density data and the Basin.
fractional-weighted sum of the predicted Vp, Vs and density
mineral properties (using the measured mineral fractions as Summary
the prediction input data). At the bottom is the model. The
properties of the carbonate mineral group are shown in We strive to map the shale reservoir properties that control
green, the properties of the clay mineral group are shown in production at the seismic scale, as the ability to identify the
red, and the silicate group properties are shown in blue. most prospective development areas is one of the keys to
Note the good match between the measurements and successfully developing regional unconventional shale
predictions, as well as the variations observed in the resource plays.
properties of the different mineral groups.
Deterministic reservoir characterization enables us to
By themselves, good correlations between the predictions estimate in-situ mineral velocities and densities for detailed
and calibration data do not guarantee a realistic earth reservoir analysis at the well log scale. In turn, this enables
model. We also need examine the cross-plots of the us to create simple models for quantitatively mapping
predicted mineral properties, and demonstrate that the spatial changes in reservoir mineralogy from 2D and 3D
predicted mineral properties are physically reasonable. For broadband seismic AVO inversion data.
this ~2,400 ft section of well data, which includes the Bone
Spring and Wolfcamp formations, we are confident that Additional value is derived from log and seismic scale
this is the case, as we observe a number of relationships mineral predictions when relationships observed in multiple
that are consistent with our prior expectations, including: North American shale plays between mineralogy and
porosity, permeability, stress anisotropy and completions
i) The estimated Vp/Vs ratio of the quartz mineral group pressure data are applicable.
is ~1.48, and ~1.86 for the carbonate mineral group.
ii) There are positive correlations between Vp and density Acknowledgements
for quartz, clay and carbonate predictions; but no
correlations between Vs and density for these mineral I am grateful to Oxy for access to well log data and for
groups. permission to publish this work.
iii) The Vp/Vs vs. Vs trends for the quartz and clay
Figure 4: [Top] Comparison of measured and predicted mineral fractions. [Middle] Comparisons of measured
and predicted rock properties. [Bottom] Depth-varying mineral properties upon which these predictions were made.
© 2017 SEG Page 4015

EDITED REFERENCES
REFERENCES
Alvarez, P., F. Bolívar, M. Di Luca, and T. Salinas, 2015, Multi-attribute rotation scheme: A tool
for reservoir property prediction from seismic inversion attributes: Interpretation, 3,
SAE9–SAE18, https://doi.org/10.1190/int-2015-0029.1.
Ball, V., M. Nasser, and O. Kolbjørnsen, 2016, Introduction to this special section: AVO
inversion: The Leading Edge, 35, 399–404, https://doi.org/10.1190/tle35050399.1.
Davie, M., Y. Zhu, O. Rehkopf, and K. Bandyopadhyay, 2012, Characterization of shale gas
reservoirs using seismic and well data at Horn River, Canada: 82nd Annual International
Meeting, SEG, Expanded Abstracts, http://doi.org/10.1190/segam2012-0273.1.
Gamero-Diaz, H., C. Miller, and R. Lewis, 2012, sCore: A classification scheme for organic
mudstones based on bulk mineralogy: AAPG Southwest Section Meeting, 40951.
Goodway, B., T. Chen, and J. Downton, 1997, Improved AVO fluid detection and lithology
discrimination using Lamé petrophysical parameters; "λρ", "μρ", & "λ/μ fluid stack",
from P and S inversions: 67th Annual International Meeting, SEG, Expanded Abstracts,
183–186, https://doi.org/10.1190/1.1885795.
Holt, R., and B. Westwood, 2016, Predicting mineralogy from elastic rock properties for low
porosity unconventional plays: CSEG Recorder, 41, 22–26.
Ødegaard, E., and P. Avseth, 2003, Interpretation of elastic inversion results using rock physics
templates: 65th Annual International Conference and Exhibition, EAGE, Extended
Abstracts, E17.
Saxena, N., G. Mavko, R. Hofmann, S. Dolan, and L. Taras Bryndzia, 2016, Mineral
substitution: separating the effects of fluids, minerals, and microstructure on P- and S-
wave velocities, Geophysics, 81, no. 2, D197–D210, https://doi.org/10.1190/geo2015-
0483.1.
Thomas, E. C., and S. J. Stieber, 1975, The distribution of shale in sandstones and its effect upon
porosity: Transactions of the SPWLA 16th Annual Logging Symposium.
van der Wal, J., and S. Stromberg, 2012, Improved workflow for evaluation of thinly bedded
sandstones – revisiting the normalised Qv equation of Juhasz: DEVEX 2012 Abstracts.
© 2017 SEG Page 4016

Playing in Tlaloc’s fields:
Trends in archaeometric prospection in the Central Mexican Highlands
Andrés G. Mejía Ramón, Department of Anthropology and Program in Human Dimensions of Natural
Resources and the Environment, The Pennsylvania State University
Summary and possible consequences their application would

have in archaeology.
When invited to participate in this session, I was
informed that archaeological presentations are
uncommon occurrences at the annual meetings of the Setting and Natural History
society. Thus, rather than planning this presentation
around the presentation of novel techniques or new Although this paper discusses archaeometric prospection in
data, I review near-surface archaeogeophysical the Central Mexican Highlands, a majority of the studies
discussed took place in the Teotihuacan Valley. Located
prospection as it is typically practiced in the Central approximately 40 km northeast of modern-day Mexico
Mexican Highlands. This will be done through a City, the Teotihuacan Valley is a semi-arid environment
series of case studies, presenting the methods consisting mainly of farmland, urban sprawl, and
employed and briefly discussing the results of each xerophytic scrub. The majority of the valley is underlain by
study. Furthermore, I will elaborate on how advances volcanic tuff mere centimeters from the surface in the
over the last two decades in geophysical technology hillsides, while up to 30-40 m deep along the alluvial plain
to which archaeology has been slow to adapt have the (Barba and Córdova Frunz 2010). The valley has been
potential to facilitate and greatly expedite both heavily modified by human agency for over 3000 years.
archaeological prospection and the process of Initially dominated by pine, oak, and cypress forests during
the Middle Holocene, the arrival of swidden farming at
excavation itself. approximately 1500-1000 BC led to massive hillside
deforestation. The loss of vegetative cover led to
subsequent massive hillside sheet erosion and valley-
Introduction bottom soil deposition (González Arqueros 2014; McClung
2015). Between approximately 200 BC – AD 550, the
There are three commonly employed geophysical Teotihuacano State---the earliest true civilization in the
categories of methods in archaeological prospection-- New World, and the largest city before the arrival of the
-magnetometry, resistivity/conductivity, and Spanish---heavily altered their environment such as in the
radiometry. As these methods are all well-established construction of large-scale bedrock-incised water
within the field of geophysics, I will forgo a management features, whose vestiges still affect the
discussion of the methods themselves, and rather modern topography (Mejía Ramón 2016). The later Aztec
focus on their application to various projects in the civilization terraced the previously-denuded hillslopes in an
Central Mexican Highlands, but especially within the attempt to reclaim marginal agricultural land (Evans 1990).
With the arrival of the Spanish, massive depopulation
Teotihuacan Valley. I will also discuss three more through disease and a change in settlement patterns dictated
novel methods---high-resolution multispectral by the colonial authority led to the collapse of the terrace
satellite imagery analysis, unmanned aerial vehicle system, leading to another episode of massive valley-wide
survey, and structure from motion---and their initial sheet and gully erosion. More recent attempts to reclaim
applications to the archaeology of the region. Given marginal agricultural land since the mid-twentieth century
differences in funding between archaeology and include the use of chisel plows which permit the creation of
geophysics, many of the tools employed by such agricultural soils, at the expense of buried archaeological
projects are several decades old, subject to remains while likely increasing long-term erosion.
technological constraint of ages past. Nevertheless,
Proyecto La Paleohidrología del Valle de Teotihuacan
recent developments in geophysical technology have
greatly increased the sensitivity of such instruments, Although I will discuss the methods and results of several
while significantly reducing the size and cost of such different archaeological projects the main focus will be on
apparatuses. I discuss a suite of these improved my project Proyecto La Paleohidrología del Valle de
technologies, elaborating on the expanded benefits Teotihuacan. This project seeks to understand Prehispanic
forms of agriculture and water management in the
© 2017 SEG Page 5110

Playing in Tlaloc’s fields
Figure 1: Identified hydraulic features near San Pedro Tlajinga
Teotihuacan Valley, especially as they relate to state typically the firstly deployed method in most
formation and the development of complex societies. The archaeogeophysial surveys. More recent studies have
project has employed magnetometric survey, electric shown that magnetometry can also be of use when looking
resistivity survey, ground penetrating radar, high-resolution for cavities and bedrock modifications such as pits and
multispectral satellite imagery, unmanned aerial vehicle canals, although the mechanisms leading to this have yet to
survey, and photogrammetric 3D reconstructions (structure be decoupled (Mejía Ramón and Barba 2017).
from motion) to identify possible buried hydraulic features
for intensive investigation and excavation (Figure 1). Since Electric resistivity compliments magnetic gradient
much of these data have not been presented to Mexico’s prospection in the identification of areas with differential
Archaeology Council as of the submission of this abstract soil moisture. These can often be interpreted as buried
(report to be submitted late May 2017), I will only briefly water deposits, intact earthen or plastered floors, as well as
comment on the preliminary results for permitting reasons. sealed stratigraphic contexts. Despite their proven
Nevertheless, the full preliminary results will be presented effectiveness, electric resistivity was of limited use in
during the annual meeting. finding subsurface hydraulic features, although this was
likely a function of the extreme aridity we encountered
Methods and Brief Preliminary Results when trying to conduct this study. Given this, I will instead
discuss previously-presented successful applications of this
As mentioned before, magnetometry, method in projects other than my own, suggesting avenues
resistivity/conductivity, and radiometry are the three main for further research.
forms of geophysical data collection in the Central Mexican
Highlands. Given its location in the volcanologically-active Ground penetrating radar is the most familiar instrument to
Trans-Mexican Volcanic Belt, a majority of architectural typical archaeologists, even if they view it as a ‘magic
features in the region are built using volcanic rock. This box’. Ground penetrating radar is the only commonly-used
lends itself very well for magnetometry and magnetic method of archaeogeophysical prospection that is capable
gradient prospection, where buried rooms and buildings can of determining the depths of features of interest.
be quickly mapped when planning excavations. Because of Nevertheless, given its sometimes-onerous post-processing,
its simplicity and limited post-processing, magnetometry is ground penetrating radar is typically not deployed for initial
© 2017 SEG Page 5111

prospection, and is limited to intensive surveys of features Arguably the most important recent advancement in
previously-identified by other methods. With the exception archaeological prospection is the increase in use of
of satellite imagery, of the discussed methods GPR was the unmanned aerial vehicles (UAVs). When coupled with
most successful method in the search for buried hydraulic structure from motion, UAVs have the capacity to deliver
structures, accurately determining the stratigraphy of all orthoimagery and topographic maps significantly more
excavated features to date. accurate than those made by more traditional methods.
Furthermore, given its low cost and ease of use, structure
While the previous three methods have long histories in from motion used in tandem with UAVs is significantly
archaeology, the final three are relatively novel to more appropriate for most investigations in non-forested
geophysics and archaeology. The Paleohidrología project environments such as the Central Mexican Highlands than
is the first to employ the use of high-resolution LiDAR, whose importance is typically overstated by
multispectral satellite imagery, unmanned aerial vehicle archaeologists. Nevertheless, recent advancements in
survey, and 3D photogrammetric reconstruction in tandem LiDAR technology, and a reduction in the size and cost of
with more traditional methods and excavation. Visual and such units are permitting the current development of
computational analysis of GeoEye-1 and IKONOS-2 inexpensive UAV LiDAR systems that could revolutionize
satellite imagery from the local wet and dry seasons proved prospection in heavily vegetated areas. The afore-discussed
extremely effective in identifying linear features throughout advancements in magnetometers also suggest that a
the valley. Excavations have confirmed that at least some possible fruitful line of engineering inquiry is the
of these are ancient canals. Unmanned aerial vehicle survey development of airborne magnetometers, which would be
and 3D photogrammetric reconstruction was employed to able to map large areas in unfavorable environments much
build a 7 km2 digital elevation model of the Southern faster and more accurately than any human could.
Middle Teotihuacan Valley, although it was surprisingly
ineffective in detecting ancient irrigation canals. Conclusions
Nevertheless, this dataset has revealed numerous
previously unidentified mounds and monumental hydraulic Though far from exhaustive, this review should
constructions that may be archaeological in nature. provide an adequate introduction to geophysicists
Application of the latter two technologies during text unfamiliar with archaeological prospection of the
excavations greatly expedited and increased precision of methods, research questions, and challenges
data recording during excavations, resulting in sub-
centimeter scale 3D reconstructions of excavated trenches
employed in the field, helping bridge the ever-present
at each stage of excavation. gap between anthropologists, archaeologists, and
geophysicist. There is a great potential for significant
Future Trends improvements in archaeometric prospection not just
in the Central Mexican Highlands, but around the
Of the three most-established forms of prospection world. This can only be accomplished through a close
discussed---resistivity/conductivity, ground penetrating relationship between investigators harking from
radar, and magnetometry---the greatest potential for future different disciplines, as well as an embrace of truly
improvement is magnetometry. Most instruments used in interdisciplinary methods and research questions.
the field are relatively large and cumbersome, and are
typically uniaxial sensors. The availability of small,
multiaxis magnetometers for well under $2,000 have the
Acknowledgements
potential to greatly increase the use of magnetometers in
archaeological projects, while allowing the detection of
subtle changes in all components of the magnetic vector.
The Proyecto La Paleohidrología del Valle de
Despite the availability of educational grants for high- Teotihuacan has been made possible by generous
resolution multispectral satellite imagery, archaeologists grants from the Stamps Family Charitable
have been slow to adopt such datasets. When they are used, Foundation, the Neukom Institute, the Claire Garber
computational analyses are rarely conducted, and their use Goodman Fund for the Ahtropological Study of
is typically limited to visual inspection and the creation of Culture at Dartmouth College, the Lucas Family
visually-appealing figures for publication. Furthermore, Fund, the Dartmouth College Class of 1996, the
archaeologists’ consideration for the temporal variable in Department of Anthropology at the Pennsylvania
satellite imagery analysis is typically lacking, with State University, and private funds from the author.
diachronic studies of imagery of the same location at
different points in time being nearly unheard of.
Satellite imagery was acquired through a generous
grant from the DigitalGlobe foundation. I would like
© 2017 SEG Page 5112

to thank Mexico’s Consejo de Arqueología and the

Ejidos de San Juan Teotihuacan and San Sebastian
Xolalpa for giving permission to this project (Permits

#403.B(4)19.2016/36/0761 and 1605). I would also
like to thank my collaborators on this project: Luis
Barba, Agustín Ortiz Butrón, Jorge Blancas, Emily
McClung (UNAM), Deborah Nichols (Dartmouth),
Oralia Cabrera (ASU), David Carballo (Boston
University), and Christian John (UC Davis). Lastly, I
would like to express my immense gratitude to the
residents of San Pedro Tlajinga and San Sebastian
Xolalpa for their hospitality and graciousness.
© 2017 SEG Page 5113

EDITED REFERENCES
REFERENCES
Barba, L., and J. L. Cordova Frunz, 2010, Materiales y energía en la arquitectura de Teotihuaca:
Universidad Nacional Autónoma de México.
Evans, S. T., 1990, The Productivity of Terrace Agriculture in Central Mexico during the Aztec Period:
Latin American Antiquity, 1, 117–132, http://dx.doi.org/10.2307/971983.
González Arqueros, M. L., 2014, Dinámica de la erosión/sedimentación en la época prehispánica y
periodo colonial. Reconstrucción de las condiciones paleoambientales en el valle de Teotihuacán
(Estado de México, México): Sc.D. Dissertation, Universidad Nacional Autónoma de México.
McClung, E., 2015, Holocene Paleoenvironment and Prehispanic Landscape Evolution in the Basin of
Mexico: Ancient Mesoamerica, 26, 375–384, https://doi.org/10.1017/S0956536115000243.
Mejía Ramón, A. G., 2016, The Paleohydrology of the Teotihuacan Valley: A multi-method approach in
the prospection of subsurface hydraulic features in Teotihuacán, Mexico: Honors A.B. thesis,
Dartmouth College.
Mejía Ramón, A. G. and L. Barba, 2017, Of Mud and Magnets: Archaeometric Prospection at the Site of
Altica: 82nd Annual Meeting of the Society for American Archaeology.
© 2017 SEG Page 5114

GPR impedance inversion for imaging the paleochannel-A case study at Mudu city site in
Suzhou, China
Yu Liu*, Zhanjie Shi and Tianxiang Yu, Zhejiang University
Summary parameter distribution is the common-midpoint velocity

analysis method (e.g., Bradford, 2008). But it provides
The survey on paleochannels is of great value for it can limited resolution (Bradford, 2009), and picking work can
help to understand former culture and geography. As a easily produce error, so it may not get accurate information
method equipped with high resolution, ground penetrating about the channel’s interface.
radar (GPR) has been widely used in archaeological
exploration. However, conventional processed profile can In seismic exploration, impedance inversion method which
only provide limited information like the shape or location has been widely used in oil industry uses well log to make
of the interface, but can’t give the distribution of physical up for seismic data’s low-frequency part, and can obtain a
properties which could help identify a paleochannel. A way full-bandwidth result (e.g., Ferguson and Margrave, 1996;
for GPR to map parameter distributions is the common- Lloyd and Margrave, 2011). GPR and seismic data analysis
midpoint velocity analysis, but it can only provide limited approaches can be mutually transferable since their
resolution. In this paper, we put forward a new scheme to similarities. Schmelzbach et al. (2012, 2015) have managed
obtain a more accurate image of the paleochannel and its to get high-resolution water content estimation by an
surrounding medium. The scheme mainly involves the adaption of seismic impedance inversion. Considering that
impedance inversion of conventional common-offset GPR archaeological work in China has rich borehole data, and
data, which uses well log to compensate GPR data and the existing impedance difference between paleochannel
finally gives a high-resolution estimation of permittivity. remains and background medium, we hope that GPR
To test this method, we did a case study at Mudu city site in impedance inversion can help solve the problems
Suzhou, China. The inversion result provides a clear image mentioned in the above paragraph. Firstly, impedance
of the channel of the ancient city’s moat. It is shown that inversion which is based on reflection amplitude can
this method is effective and feasible for detection of provide high resolution, and its result would be more
paleochannels. accurate with compensation from well data. Secondly,
impedance value has a direct transformation relationship
Introduction with permittivity under normal assumptions, so that the
inversion result can be connected with soil attribute. This
The study of paleochannels not only has a significance in could facilitate archaeologists’ identification of the
hydrology, engineering and mineral exploration, but also paleochannel.
can help archaeologists understand ancient culture and
geography (e.g., Karmanov et al., 2013). Some ancient In this paper, we present a case study using impedance
channels are of great archaeological value since they could inversion of GPR data to detect the moat of Mudu Ancient
help reveal historical truths. For the discrimination of City, for the purpose to prove this method’s validity and
paleochannels that are buried underground, drilling or feasibility.
geophysics methods would be needed. In comparison with
frequently-used borehole sampling method, geophysics Field Description
survey is non-destructive and more efficient, and has the
ability to provide more lateral information (e.g., Holzschuh, Mudu Ancient City, located at current Suzhou of Jiangsu
2002). Province, China (Figure 1), dates back to Spring and
Autumn Period. It is preliminarily considered as a capital
Ground penetrating radar (GPR) is a non-invasive, high- city at that time, for many relevant things like tombs, jade
efficiency and high-resolution geophysics method, which cellars were discovered nearby. This large-scale and high-
has been extensively applied in archaeological exploration grade ancient city was listed as one of China’s top ten
(e.g., LIN et al., 2011; Zhao et al., 2015; Shi et al., 2015). archaeological discoveries in 2010. Plenty of surveys and
However, in most GPR detection cases at present, the data excavations show that Mudu Ancient City’s shape was
are just routinely processed. The interface of subsurface irregular, and its walls distributed roughly along the edge of
formation is provided, but permittivity which is important basin (Zhang, 2002). The north city wall is located at
to identify the properties of subsurface formation cannot be Wufeng village area. An archaeological team from Chinese
acquired. This is insufficient for detection of paleochannels. Academy of Social Sciences employed drilling method in
A common way for GPR to estimate soil physical 2011 and confirmed that there existed a moat outside the
© 2017 SEG Page 5115

GPR impedance inversion for imaging the paleochannel
Figure 1: Location map of Mudu Ancient City
wall(north and east side) with a similar extension. At a Stage 1: GPR Data Processing
known excavation site, average depth of the channel bed is ① GPR data preprocessing. This includes time zero
about 1.5m, and large amounts of channel-fill deposits exist. corrections, DC removal, dewow, etc.
Raw soil could be detected below the channel’s bottom ② Conventional GPR data processing. This includes
interface. However, most parts of the moat have been amplitude recovery, background removal, frequency filter,
buried by overlying soil. The study area in this paper is deconvolution, and migration, etc.
located at a buried moat area outside the Wufeng section of ③Sparse spike deconvolution, to estimate the reflectivity.
city walls. The area is flat, with little interference above the
④ Time-depth conversion using the reference velocity.
ground.
Then chop the data under 2m, and shift back the rest to
time domain. The result is called as R.
Data Acquisition
Figure 2 shows the result after above processing of GPR
We performed a 25m-long GPR survey line perpendicular
data. Under the assumption that the reflectivity series
to the ancient city wall in 2016. In consideration of the high
associated with a layered Earth model can be represented as
soil moisture in southern China, the 100MHz antennas
a series of sparse spikes, the third step inverts a GPR trace
were selected to lighten signal’s attenuation. We measured
to reflectivity series. Sparse spike deconvolution can
the GPR data in common-offset mode by moving a pairs of
provide the times and sizes of the largest reflectors and
antennas with a constant separation of 1m and 0.1m trace
effectively recover the lost high-frequency part of the
spacing. Sampling interval was 0.4ns and 64 stacks were
reflectivity. However, some low-frequency information
taken at each trace. The reference velocity information was
which is essential for creating the character or scale of the
got by means of the reflection wave method. Following the
impedance is still missing. So we need supplementary data
GPR survey, we drilled three cores 2 meters deep at 5m,
from the wells.
10m and 25m of the survey line using a Luoyang Spade (a
very popular and efficient drilling tool in Chinese
archaeological work), and all detected raw soil under the
paleochannel’s bottom interface. A test instrument was
used to measure the permittivity at every 0.1m of the soil
core.
Inversion Scenarios
Our inversion workflow consists of three parts, namely,

GPR data processing, well data processing and impedance
inversion. The first one is performed by professional
processing software assisted by some MATLAB programs. Figure 2: Reflectivity profile after GPR data processing
Other two are all performed on MATLAB.
© 2017 SEG Page 5116

GPR impedance inversion for imaging the paleochannel
Stage 2: Well Data Processing green boxes represent the locations of topside raw soil in
① Shift the permittivity on the well log to impedance the boreholes, and the black dotted line represents the
values, and for each well interpolate the data to get the paleochannel’s bottom interface.
same number of samples as a GPR trace.
②Apply a two-dimensional cubic spline interpolation by
the data of three wells, and get a low-frequency impedance
volume.
③ Perform a depth-time conversion to the impedance
volume, and get the result 𝑍𝑤 with the same size with R.
Figure 3 shows the result after above processing of well

data. Since the number of acquired sample points in the
well data is much less than that in GPR data when the depth
is fixed, interpolated well data in time domain can be
regarded as low-frequency information. Thus we can merge Figure 4: Result of impedance inversion
two volumes to get a full-bandwidth result.
Figure 3: Low-frequency impedance volume Figure 5: Analysis result after smoothing
Stage 3: Impedance Inversion Discussion of Results/Conclusions

①Apply a band-limited integration filter to R, and compute
recursively to get the impedance volume 𝑍1 ; The inversion result gives a better view of location and
②Compute 𝑍𝑏 which is the linear trend of 𝑍𝑤 and remove shape of the paleochannel than conventional processed data.
it from 𝑍𝑤 . The result is named 𝑍2 ; The channel bed’s depth and channel’s width correspond
③Compute the spectrums of 𝑍1 and 𝑍2 . with the nearby excavation results. By GPR impedance
④Determine a scalar to match the spectrum of 𝑍1 to that of inversion, high-resolution image of distribution of channel-
𝑍2 . Apply a low-pass filter to spectrum of 𝑍2 and add it to fill deposit or raw soil can be provided, which is very
spectrum of 𝑍1 . beneficial to archaeological work.
⑤Inverse Fourier Transform the result in the fourth step,
and add 𝑍𝑏 to it. We conclude that GPR impedance inversion offers an
effective and feasible means to the detection of
⑥Perform a time-depth conversion to the result in the fifth
paleochannels. It has the ability to enhance imaging and
step, and optionally shift impedance values to permittivity
characterization of such historic site. With its high-
values.
resolution results, archaeological excavation plans can be
optimized. But for very inhomogeneous soil rather than soil
Figure 4 shows the final inversion result. By merging with
with mild lateral property variations, this method may need
well data, the low-frequency components missing from
to be adapted.
GPR data are compensated, and we get a full-bandwidth
impedance distribution. After shifting to permittivity
Acknowledgements
values, we can try to recognize different types of soil. For
instance, the channel-fill deposit shows larger permittivity
The authors gratefully acknowledge the support of Chinese
value than raw soil in the well data, so the interface of them
Academy of Social Sciences and Suzhou Institute of
could be identified. Considering that the soil showed mild
Archaeology during data acquisition and archaeological
lateral property variations, we apply a horizontal five-point
interpretation.
smoothing filter (Figure 5) to make it easy to analyze. The
© 2017 SEG Page 5117

EDITED REFERENCES
REFERENCES
Bradford, J. H., 2008, Measuring water content heterogeneity using multifold GPR with reflection
tomography: Vadose Zone Journal, 7, 184–193, http://doi.org/10.2136/vzj2006.0160.
Bradford, J. H., W. P. Clement, and W. Barrash, 2009, Estimating porosity with ground-penetrating radar
reflection tomography: A controlled 3-D experiment at the Boise Hydrogeophysical Research
Site: Water Resources Research, 45, http://doi.org/10.1029/2008WR006960.
Ferguson, R. J., and G. F. Margrave, 1996, A simple algorithm for band-limited impedance inversion:
CREWES Research Report, No. 8.
Holzschuh, J., 2002, Low-cost geophysical investigations of a paleochannel aquifer in the Eastern
Goldfields, Western Australia: Geophysics, 67, 690–700, http://doi.org/10.1190/1.1484512.
Karmanov, V. N., A. V. Chernov, N. E. Zaretskaya, A. V. Panin, and A. V. Volokitin, 2013,
Paleochannel studies in archaeology: The case of the Vychegda river, Northeastern European
Russia: Archaeology, Ethnology and Anthropology of Eurasia, 41, 83–93,
https://doi.org/10.1016/j.aeae.2013.11.008.
Lin, J., G. Tian, B.-B. Wang, D.-S. Xu, J.-H. Liu, and W.-K. Zhao, 2011, Integrated geophysical survey
for palaeo-drainage system at Liangzhu sites: Journal of Zhejiang University (Engineering
Science), 45, 954–960, http://doi.org/10.3785/j.
Lloyd, H. J. E., and G. F. Margrave, 2011, Bandlimited impedance inversion: using well logs to fill low
frequency information in a non-homogenous model: CREWES Research Report, No. 23.
Schmelzbach, C., J. Tronicke, and P. Dietrich, 2012, High-resolution water content estimation from
surface-based ground-penetrating radar reflection data by impedance inversion: Water Resources
Research, 48, http://doi.org/10.1029/2012WR011955.
Shi, Z., G. Tian, R. W. Hobbs, H. Wo, J. Lin, L. Wu, and H. Liu, 2015, Magnetic gradient and ground
penetrating radar prospecting of buried earthen archaeological remains at the Qocho City site in
Turpan, China: Near Surface Geophysics, 13, 477–484, http://doi.org/10.3997/1873-
0604.2015033.
Zhao, W., E. Forte, G. Tian, M. Pipan, Y. Wang, X. Li, Z. Shi, and H. Liu, 2015, Advances in GPR data
acquisition and analysis for archaeology: Geophysical Journal International, 202, 62–71,
http://doi.org/10.1093/gji/ggv121.
Zhang, Z., 2002, Investigation and excavation of large city at spring and autumn period in Suzhou:
Journal of Suzhou Railway Teachers College (Social Science Edition), 19, 93–96.
Zeng, Z. F., X. Chen, J. Li, L.-N. Chen, Q. Lu, and F.-S. Liu, 2015, Recursive impedance inversion of
ground-penetrating radar data in stochastic media: Applied Geophysics, 12, 615–625,
http://doi.org/10.1007/s11770-015-0514-0.
© 2017 SEG Page 5118

The HVSR technique from array data, speeding up mapping of paleo-surfaces and buried
remains. The case of the Bronze-Age site of Pilastri (Italy)
Samuel Bignardi*, Nasser Abu Zeid, Erica Corradini and Giovanni Santarato, University of Ferrara, Italy
Summary water taken from a nearby river. Inside, they were

artificially flooded and houses were built on a wood
In the framework of a geophysical survey at the “Pilastri platform. The Emilian Plain is composed of a thick
Terramara” Middle Bronze Age archeological site sequence of alluvial sediments, mainly clayey-silt and its
(Northern Italy), which aimed at using cost-effective landscape is characterized by the presence of slightly
approaches to investigate a 12.000 m2 portion of land, we elevated and elongated features, composed of sand, related
verified that in this site the most commonly used GPR, to paleo-riverbeds that constituted the ideal place to build
magnetometry and ERT geophysical methods were the "terramare" settlements and often, in later times, were
hampered by specific unfavorable conditions. Since the re-occupied by Romans.
main purpose of the survey was to obtain information on
ancient walking surfaces, we speculated that compaction,
caused by the trampling of human activity over centuries,
increased both density and seismic velocity of these paleo-
surfaces. Assuming that such compacted layers still lie
embedded in the shallow subsurface we successfully used
the Horizontal-to-Vertical Spectral Ratio of microtremors,
popular as HVSR, as a mapping tool. HVSR technique is
seldom used in archaeological studies. The curve obtained
by this method shows peaks and features that can be
associated to the elastic impedance contrasts of a layered
subsurface. Results from a previous single-station HVSR
survey carried out on the whole area and consisted of 67
measurement points, allowed to obtain an indication about
acoustic impedance contrasts and to spot few anomalies.
The preliminary, yet promising result however, needed
further acquisitions in order to achieve a better spatial
sampling. We then tested microtremors acquisition using a
linear array of seismic prospection geophones (4.5 Hz Figure 1: Aerial view and location of the Pilastri arcaeological site.
proper frequency) which have a worse response with
respect the short period sensor used in single-station Recent studies (Nizzo, 2014; Nizzo et al., 2015)
configuration. Yet the use of the array allowed for a hypothesized that the Pilastri settlement could actually
comparable mapping of both paleo-surfaces and anomalies extend outside the originally established bordering (Balasso
and data collection was performed at 24 locations and Michelini, 2013, personal communication). Therefore,
simultaneously, so greatly accelerating the field work. In the main purpose of the geophysical investigation was to
the following the two approaches are compared and an use a cost-effective tool for the recognition and localization
anomaly is highlighted which was located outside the of paleo-surfaces beyond the eastern limit of the main
hypothesized Bronze Age settlement area. Once excavated, settlement, over an area of about 12,000 square meters.
such feature revealed to be a layer of roman age remains.
Further, we show our best GPR section over the same The technologies available to our research group and
anomaly. An indication was present in the GPR section but mostly encountered in archaeological studies are
this became clear only afterward. magnetometry, Electrical Resistivity Tomography (ERT),
and GPR. Unfortunately however, the Pilastri site lies in
Introduction part under the iron structure of greenhouse cultivations and
is surrounded by houses at the eastern and northern edges.
The archaeological site of Pilastri is part of the These circumstances made the magnetometry survey
demographic phenomenon of the Middle Bronze Age useless. Further, the nature of the target we wanted to find,
popular as “terramare” (1700/1650-1350/1300 BC), in the i.e. thin compacted silty-clayey layers embedded in less
Emilian Po plain (Northern Italy, Figure 1). The typical compacted material of the same lithological nature, makes
structure of these settlements was quadrangular in shape, it difficult to apply the ERT method (Santarato, 2003).
surrounded by a terrigenous embankment and a ditch Once considered the area to cover, the costs and the
(Chierici, 1871; Desittere, 1997), and often filled with working time necessary for the data collection, such a
© 2017 SEG Page 5119

HVSR technique in array configuration: a tool for mapping paleo-surfaces and anomalies
target turns to be well beyond the resolving power of any

reasonable choice of electrode spacing. Finally, the
prevailing clayey texture of the sediments severely

hampered the use of the GPR. Our tests showed that when
using a 400 MHz antenna the investigation depth was
severely reduced, while using a 100 MHz one resulted in an
insufficient resolving power. Besides these difficulties, a
proper acquisition of both ERT and GPR data to map such
an extended area requires a dense grid of profiles, which
would make the survey exceedingly time consuming.
Therefore, we decided to test a method which is much more
common in geology rather than in archaeology: the HVSR.
The HVSR method is based on the evaluation of the

resonance frequencies due to the presence of layers with
increasing acoustic impedance. The rationale for using this
approach in such a context is primarily the assumption that
the paleo-surfaces were stiffened due to trampling of
human activity over centuries of occupation. Secondly,
HVSR is able to map subsurface elastic variations spanned
over short distances (Aki and Richards, 2002, Bignardi et
al., 2014, Bignardi et al., 2016), hence allowing capturing
such variations by acquiring data over an optimized grid of
locations. HVSR applications are scarcely documented in
archaeology and few examples exist whose scope is
different from the present application, e.g.: Castellaro et al.
(2008), Bottari et al. (2012), Wilken et al. (2015),
Obradovic et al. (2015). In previous work, Abu Zeid et al.
(2016, 2017) showed 2D vertical slices, derived from the
3D volume built by the single-station data as compared
with the 2D ERT. In the present work we show a
comparison of 2D sections (Figure 2) obtained both from
single-station and from an array configuration. Despite the
fact that prospection designed geophones (4.5 Hz) were
used, which have a worse low frequency response, the
array achieved results comparable to those of the single-
station. Further, the array approach acquired 24 receivers
simultaneously so that the result was achieved 24 times
faster and using low-cost equipment.
Figure 2: Section obtained by placing normalized HVSR
Furthermore, we show the portion of our best GPR section curves side by side. a) section obtained from 22 single-
acquired on this site (Figure 3), which lies exactly over an station measurements (Geophone with 2Hz proper
anomaly that a subsequent excavation proved to be a thin frequency). b) section obtained by a measurement
layer of Roman remains, consisting for the most of brick comprising the acquisition at 24 geophones (4 Hz proper
fragments. As it is clear by inspecting the section b) in frequency) simultaneously. HVSR amplitudes are
Figure 2, HVSR was successful in highlighting the buried normalized. Label “A” indicates an old, re-filled excavation
target, while the GPR result became clear only after the while “B” highlights the excavation that we suggested,
excavation. based on these results, where Roman remainings were
found (see figure 3). White lines highlight possible paleo-
The consideration to be made is that HVSR may represent a surfaces, while short black lines indicate local maxima of
useful, non-invasive tool for mapping very shallow targets, HVSR curves.
especially in array configuration, and it is worth to be .
included as a candidate member of the family of
Method
geophysical methods useful in archaeo-geophysics.
© 2017 SEG Page 5120

The HVSR method, (Nogoshi and Igarashi, 1971;

Nakamura 1989, 2000), is based on the acquisition of the
three components of ambient seismic noise that is induced

by different sources that can be both natural and human
generated. The ratio between the power spectra of the
horizontal and vertical components is
calculated by
√
(1)
and the peaks of the resulting curve can be directly related

to elastic impedance contrasts in the subsurface. Typically,
the subsurface is assumed to be well described by a low
shear wave velocity ( ) layer of thickness H, over a fast
half space. In such a case the resonance frequencies are
known
(2)
If the soft layer contains minor impedance contrasts, the

HVSR curve will show multiple peaks. In the latter case,
linking the experimental H/V curves to the layered
subsurface requires an inversion algorithm (Castellaro et
al., 2005, Herak, 2008; Mantovani et al., 2015; Bignardi et
al., 2016).
Here we discuss a 2D section obtained by placing side by

side the HVSR curves obtained for a set of aligned
locations. In such representation the y-axis is frequency and
for sake of interpretation needs to be transformed into
depth. Since we are interested in the very shallow Figure 3: Portion of the GPR survey performed with a 400 MHz
subsurface, down to 2 m depth, to translate the frequency antenna before the excavation. A weak anomaly produced by the
axis to a depth, inversion is actually not needed and it is buried remainings is present, however it could not be spotted on
the basis on the lone GPR. Big and small squared highlight the
sufficient a simplified approach. Indeed, the site is excavation location and the radar signature of the Roman
characterized by a thick sedimentary cover which results in remaining.
a main resonance peak around 0.8 Hz. This peak dominates
over the entire HVSR curve, so that in such a context
available inversion algorithms usually give poor results for [ ] , (3)
frequencies higher than 10 Hz, mainly because the
inversion would be dominated by the deep subsurface where is the selected resonance frequency, and
structure. Moreover, because of the weak impedance
m/s. Figure 2a and 2b show the HVSR section
contrast represented by the wanted paleo-surfaces any
built from both single-station and from array measurements
inversion procedure would suffer from severe equivalence
respectively. For the single-station measurement 22 curves
issues.
were merged. The equipment used to acquire the single-
station dataset comprised a 3D short-period geophone (2
Therefore, as the shallow lithology of the site is almost
Hz) connected to a M.A.E. Vibralog data-logger. For the
homogeneous with an average of the shallowest layers
array measurement we used 24 three component geophones
known by a nearby borehole (30 m deep), we decided to (4.5 Hz) connected to a seismograph purposely built by our
apply the simplified approach proposed by Ibs Von Seht research group.
and Wohlemberg (1999), equation 3, which accounts for
the increasing of Vs with depth due to the increasing
Since we were mainly interested in the shallow subsurface
confinement pressure which, in turn, involves the investigation of the frequency
range above 10 Hz, microtremors recordings 15 minutes
© 2017 SEG Page 5121

long were sufficient. The two acquisitions were performed so to create an elastic impedance contrast detectable by
with different location spacing. Receivers are highlighted investigating the HVSR curve at sufficiently high
by red circles on top of the HVSR sections of figure 2. frequencies. Indeed, such paleo-surfaces were found in
Further, considerations about the sections indicate that connection with small local maxima of the curves which in
despite the apparent graphical difference they look coherent some cases resulted nicely aligned to form smooth surfaces.
in their main features. Some differences however, were to Further, highlighting in the interpolated sections the depths
some extent expected, basically because of three reasons: corresponding to these local maxima allowed following at
 Difference of receivers employed which have glance the surfaces. Finally, presence and estimated depth
different response both in term of proper of paleo-surfaces were afterwards confirmed by direct
frequency and amplitude response. excavations.
 Difference in acquisition times. Single station was
performed one location at a time, while the array
recorded simultaneous signals. Further, the two
surveys were performed at different times.
 Stochastic nature of the seismic noise and the
consequent different “illumination” of the
subsurface.
An anomaly “A” could be recognized as coincident, both in

position and depth, with a previously excavated and then
re-filled site. Square “B” highlights an anomaly that we
suggested to excavate, which turned out to be caused by the
presence of a thin layer of Roman earthenware and brick
fragments at a depth of 70 cm (Figure 4). White lines
drown based on section in figure 2b, and superimposed on
Figure 4: Excavation performed after the geophysical survey. A
both sections, highlight possible surfaces of interest. thin layer of Roman remainings was found at the exact place and
Surface “C” lying roughly at a depth of 40 cm is most depth obtained by the HVSR investigation.
certainly coincident with plough base. Surfaces “D” are
probably discontinuous portions of a paleo-surface. They We showed that equivalent results may be obtained both
lie at a depth of about 70 cm, the same depth of the using single-station and array configurations. Despite that
discovered Roman remains. 4.5 Hz low-cost prospection geophones were used the array
performed nicely and allowed recording 24 stations
Figure 3 shows a detail (10 meters centered at x=125 m, simultaneously, so greatly reducing acquisition times. To
corresponding to anomaly “B” of figure 2), of the best GPR the best of our knowledge, this configuration is used for the
result we were able to achieve over the same profile using a very first time in archaeological context. The success of
400 MHz antenna. The full radargram looks as the shown this field application shows that the proposed HVSR
image for its entire length (280 m). As such no indication approach was capable of investigating interfaces of
was obtainable only on the basis of GPR. In fact, the archaeological interest which spread over vast areas, and
anomaly signature is very weak so that its presence was could be routinely used to pin ancient shallow paleo-
identified only after that we knew what to look for. Finally, surfaces whenever other well-documented geophysical
in figure 4 we show a picture of the open excavation with methods can’t be used and/or when a low-cost
the Roman remains. reconnaissance survey is requested.
Conclusions Aknowledgments
We presented and discussed a successful application of the The authors would like to thank the Superintendence for
HVSR geophysical method at the archaeological site of Archaeological Heritage of Emilia Romagna, the Bondeno
Pilastri (Northern Italy). Geological setting and town Municipality, Associazione Bondeno Cultura, Gruppo
urbanization conditions of the investigated area do not Archeologico di Bondeno, the community of Pilastri town
allow a successful use of the geophysical methods most and the landlord of the site. Sincere thanks are due to the
commonly employed in archaeology. Therefore, we Archaeological Firm ‘P.E.T.R.A.’ for their kindness in
decided to test whether HVSR can be used to gain providing relevant pictures related to present and past
information about the local subsurface, and in particular, excavations. Authors would finally to thank the site-
for detecting paleo-surfaces. The hypothesis was that these director Valentino Nizzo his valuable suggestions and for
paleo-surfaces were compacted and stiffened by trampling, the professional coordination.
© 2017 SEG Page 5122

EDITED REFERENCES
REFERENCES
Abu Zeid, N., E. Corradini, S. Bignardi, and G. Santarato, 2016, Unusual geophysical techniques in
archaeology–HVSR and induced polarization, a case history: 78th Annual International
Conference and Exhibition, EAGE, Extended Abstracts, NSAG-2016,
https://doi.org/10.3997/2214-4609.201602027.
Abu Zeid, N., E. Corradini, S. Bignardi, V. Nizzo, and G. Santarato, 2017, The passive seismic technique
‘HVSR’ as a reconnaissance tool for mapping paleo-soils: The case of the Pilastri archaeological
site, northern Italy: Archaeological Prospection, https://doi.org/10.1002/arp.1568.
Aki, K., and P. G. Richards, 2002, Quantitative Seismology: 2nd ed.: University Science Books.
Bignardi, S., G. Santarato, and N. Abu Zeid, 2014, Thickness variations in layered subsurface models–
Effects on simulated MASW: 76th Annual International Conference and Exhibition, EAGE,
Extended Abstracts, WS6Â P04, https://doi.org/10.3997/2214-4609.20140540.
Bignardi, S., A. Mantovani, and N. Abu Zeid, 2016, OpenHVSR: imaging the subsurface 2D/3D elastic
properties through multiple HVSR modeling and inversion: Computers and Geosciences, 93,
103–113, https://doi.org/10.1016/j.cageo.2016.05.009.
Bottari, C., S. Urbini, M. Bianca, M. D'Amico, M. Marchetti, and F. Pizzolo, 2012, Buried archaeological
remains connected to the Greek-Roman harbor at Tindari (north-east Sicily): Results from
geomorphological and geophysical investigations: Annals of Geophysics, 55, 223–234.
Castellaro, S., F. Mulargia, and L. Bianconi, 2005, Passive Seismic Stratigraphy: A new efficient, fast and
economic technique: Journal of Geotechnical and Environmental Geology, 3, 51–77.
Castellaro, S., S. Imposa, F. Barone, F. Chiavetta, S. Gresta, and F. Mulargia, 2008, Georadar and passive
seismic survey in the Roman Amphitheatre of Catania (Sicily): Journal of Cultural Heritage, 9,
357–366, https://doi.org/10.1016/j.culher.2008.03.004.
Chierici, G., 1871, Le antichità preromane della provincial di Reggio nell'Emilia. Reggio Emilia (in
Italian): Tip. Stefano Calderini.
Desittère, M., 1997, La scoperta delle terramare e gli inizi degli studi di preistoria in Italia nell'età
dell'evoluzionismo e del positivismo, in M. Bernabò-Brea, A. Cardarelli, and M. Cremaschi, eds.,
Le Terramare, la più antica civiltà padana, Electa Milan, 59–64 (in Italian)
Herak, M., 2008, ModelHVSR–A Matlab® tool to model horizontal-to-vertical spectral ratio of ambient
noise: Computers and Geosciences, 34, 1514–1526, https://doi.org/10.1016/j.cageo.2007.07.009.
Ibs-von Seht. M., and J. Wohlenberg, 1999, Microtremor measurements used to map thickness of soft
sediments: Bulletin of the Seismological Society of America, 8, 250–259.
Nakamura, Y., 1989, A method for dynamic characteristics estimation of subsurface using microtremor
on the ground surface: Quarterly Report of Railway Technical Research Institute, 30, 25–33.
Nakamura, Y., 2000, Clear identification of fundamental idea of Nakamura’s technique and its
applications: 12th World Conference on Earthquake Engineering: New Zealand Society for
Earthquake Engineering, New Zealand.
Nizzo, V., 2014, Memoria and Terremoto, scavo della terramara di Pilastri (Bondeno, FE): Forma Urbis
XIX, 10, 51–52 (in Italian).
Nizzo, V., A. Balasso, S. Bergamini, M. Cupitò, L. Dal Fiume, M. Marchesini, S. Marvelli, P. Michelini,
G. Osti, M. Pirani, S. Tassi, and M. Vidale, 2015, Lo scavo della ‘Terramara’ di Pilastri
(Bondeno, FE): Storia di un’esperienza condivisa, tra Memoria and Terremoto: Forma Urbis XX,
2, 42–56 (In Italian).
© 2017 SEG Page 5123

Nogoshi, M., and T. Igarashi, 1971, On the amplitude characteristics of microtremor. Part 2: Journal of
the Seismological Society of Japan, 24, 26–40 (in Japanese with English abstract),
https://doi.org/10.4294/zisin1948.24.1_26.
Obradovic, M., N. Abu Zeid, S. Bignardi, M. Bolognesi, M. Peresani, P. Russo, and G. Santarato, 2015,
High Resolution Geophysical and Topographical Surveys for the Characterisation of Fumane
Cave Prehistoric Site, Italy: 77th Annual International Conference and Exhibition, EAGE,
Extended Abstracts, https://doi.org/10.3997/2214-4609.201413676.
Santarato, G., 2003, Le indagini preventive: la diagnostic non invasiva, in A. Minelli, A., C. Peretto, eds.,
Metodologie per lo scavo archaeologico: Centro Europeo di ricerche preistoriche, Poligrafica,
43–60 (in Italian).
Wilken, D., T. Wunderlich, B. Majchczack, J. Andersen, and W. Rabbel, 2015, Rayleigh-wave resonance
analysis: A methodological test on a Viking age pit house: Journal of Cultural Heritage, 9, 357–
366.
© 2017 SEG Page 5124

Exploring the paleolithic cave of Fumane (Italy): Geophysical methods as planning tool for
archaeology
Nasser Abu Zeid, Samuel Bignardi*,Giovanni Santarato and Marco Peresani, University of Ferrara, Italy
Summary
Geophysical techniques are seldom used for the

investigation of paleolithic archeological sites, mostly due
to their sedimentary nature and the almost complete
absence of architectural remains that can result in clear
geophysical anomalies. The Fumane cave (Italy) is
considered one of the most important sites in Europe.
Recent investigations and studies carried out by the
archaeological group of the Dep. Of Humanities
(University of Ferrara) with the enrollment of the
Department of Engineering of the same University allowed
for the reconstruction of the 3D model of the Fumane Cave.
This offered an opportunity for the applied geophysics
group of the same university to assess the capability of
Electrical Resistivity Tomography to retrieve subsurface
information of archaeological interest. As primary goals,
the study aimed at the creation of a three-dimensional
resistivity model of the subsurface, to infer the nature of the
sedimentary infill, so enabling a better understanding of the
depositional processes involved in the formation of this
important archaeological deposit and at the same time, for
planning long-term field-investigations and to locate areas
interesting for excavation. Moreover a Ground penetrating
radar and a Nakamura survey were attempted in an
integrate approach to test whether the combined
methodologies could provide further insight about the
subsurface. With respect to previous work, we present new
and improved results both on ERT, GPR and HVSR. In
particular, the HVSR test was accomplished to gain
information about the maximum thickness of the deposit
and to see if other impedance contrast discontinuities are
present as was anticipated by previous archaeological
studies.
Introduction
Ground surface geophysical prospection methods have

been used extensively in archaeology for non-invasive
exploration of shallow-depth targets. Different techniques
are of course sensible to different physical properties so
that, when used together in an integrated geophysical
survey, they allow reaching a deeper insight about position
and extension of anomalous bodies present in the
subsurface. Figure 1: selected 3D resistivity slices of the 3D ERT
inversion model. Blu: low resistivity volumes associated
In a previous study (Obradovic et al. 2015) presented 2D with fine grained sediments. Red: high resistivity volumes
Electrical Resistivity Tomography (ERT) results of this associated with troughs filled with drained coarse-grained
paleolithic deposit (Paresani et al., 2008; Peresani, 2012). sediemnts.
© 2017 SEG Page 5125

Exploring the paleolithic cave of Fumane (Italy) using Geophysical methods
They showed that the methodology is of high value to presence of reflectors due to ancient rock falls, buried
search for sediment accumulations of different textures. boulders or other more regular geometries.
Further they showed that resonance peaks obtained by The proposed integrated methodology is intended to
Horizontal over Vertical Signal Ratio (HVSR: Nakamura, achieve the following objectives in the long-term period: 1)
1989, 2000) can be observed and empirically correlated to to execute a high resolution ERT survey to investigate the
possible presence of subsurface acoustic impedance shallowest part of the deposit and obtain more insight on its
contrasts. In this work the ERT data was further elaborated geometry and texture distribution; 2) to localize zones of
so attaining a 3D resistivity model. Further, Ground greatest archaeological interest and to identify possible
Penetrating Radar (GPR) results are shown for the very voids and/or channels; 3) to use microtremors to infer
first time. While GPR tests were carried out at specific discontinuities of acoustic impedance present within and/or
locations to support the ERT model, some HVSR at the base of the deposit; 4) to generate an high resolution
measurements were undertaken to possibly obtain and geo-referenced three-dimensional model of the
information about the maximum depth of the deposit. subsurface of the cave to be combined with previously
While in previous work (Obradovic et al., 2015) HVSR acquired 3D laser scanner and photogrammetry data in a
curves were only qualitatively discussed, here a shear wave whole 3D model of the cave (Bolognesi et al., 2015). In this
velocity model is obtained by inversion. This integrated way, a great visual and communicative impact is achieved
geophysical survey was undertaken in order to achieve a on a wide professional and non-professional public.
better understanding of the evolution of this cave, as well
as to detect hidden sedimentary structures of interest. The
purpose of such investigation is mainly to give indications
to be used for planning the archaeological fieldwork
The achieved result paves the strategy for the developments
of a virtual archiving system were to collect all the
available and future knowledge of this important site. Such
a system should be of major help to researchers for
planning the archaeological fieldwork and may constitute
the first step towards enhancing its visibility to the broader
public.
Methods
The geophysical survey was conducted mainly by means of

the ERT, while a GPR and a HVSR survey were performed
to investigate other physical properties to put light on the
nature of observed resistivity anomalies.
While GPR is commonly used in archaeology, its

application in paleolithic caves is very rare, perhaps
because of the absence of those targets that are quite easily
detected in archaeological sites. Further, despite the fact
that HVSR is seldom used in such a context, successful
applications of the method in archaeology exist (Abu Zeid
et al. 2016, 2017), so that we decided to investigate the use
of these techniques to cross-check and possibly integrate
ERT results, therefore producing more reliable result. The
ERT has the power to reconstruct the vertical and
horizontal electrical stratigraphy (i.e. resistivity), while the
HVSR spectral ratio helps in inferring the presence of
lithological discontinuities characterized by different
seismic velocities. The use of the HVSR method aimed to Figure 2: Two selected radargrams. Radargram a) was recorded
infer the depth to the hard bedrock and to identify possible along the line GPR1 (figure 1) on the top of the deposit with a
shallow elastic reflectors hence extending the investigation 400 MHz antenna. Real amplitude is shown. Radargram b)
depth of the ERT survey, confined to the shallowest 3-4 m, highlighted as GPR2 was recorded vertically, along the deposit
to the whole thickness of the deposit, estimated around 5-6 wall of figure 3c using a 200 MHz antenna. Total energy view is
m. Finally, GPR was used mainly to investigate the shown.
© 2017 SEG Page 5126

can be associated to medium large blocks, probably

The ERT survey comprised 11 profiles in total, one-meter detached from the ceiling, or alternatively, to dense
spaced and covering the main excavated area (numbered accumulation of frost shattered stones. Medium and high
lines in Figure 1). The electrode spacing was set to 0.5 m resistivity volumes predominate the subsurface of the
and data were collected using the ABEM SAS 4000 southern part of the deposit at the cave entrance where
Multichannel georesistivity-meter. Profiles length was sediments are expected to be less humid and are also
variable between 6.5 and 12 meters. Since we wanted to expected to contain more frost-shattered stones of medium-
achieve the maximum quality result, the measurements large size.
were collected using both the Pole-Pole and Wenner-
Schlumberger electrode configurations, to take advantage
of the greater depth of investigation offered by the former,
while retaining the enhanced resolving power of the
second, especially in the shallowest 2 meters. Apparent
resistivity data retained after preprocessing was inverted to
obtain the best estimate of the true subsurface resistivity.
To this end ERTLab 3D code was used. The code can
handle rough topography so that the free surface
represented by the two excavated sides of the deposit was
handled by imposing suitable boundary conditions (Morelli
and LaBrecque, 1996).
Concerning the passive seismic data, it was collected using

a three component 2 Hz electromagnetic seismometer
connected to a 24 bit seismograph (model Vibralog M.A.E.
Italy). The equipment was positioned at three strategic
points (Circles in Fig. 1a) with the seismometer oriented
northward, which is also parallel to the cave walls. Seismic
noise data were recorded continuously for one hour at 250
Hz sampling rate. The analysis performed using the HVSR
method as implemented in the Geopsy (Di Giulio et al.,
2006, Wathelet et al., 2008) was qualitatively discussed in
(Obradovic et al., 2015) In this context, HVSR curves
were inverted using OpenHVSR (Bignardi et al., 2016).
Finally, Radar scans were acquired using a 200 MHz

antenna connected to the acquisition unit model RIS PLUS
by IDS S.p.A. (Pisa, Italy). As the surface of the
archaeological site could not be walked, acquisition took
place by manually moving the antenna over the surface of
the deposit.
Results
In Figure 1b-c slices at significant levels of the 3D ERT

model are shown. The shallowest shows few anomalies
with high spatial resistivity variations, especially along the
eastern portion of the cave. The features of main interest
are the zones with resistivity values lower than 100 Ohm.m
(ERT lines 1 to 4 in Figure 1a), which extend to a depth of
about 3 to 4 meters below the walking surface. Such zones Figure 3: a) Result of HVSR inversion. Fitting between data and
are most probably the richest ones in fine-grained the synthetic HVSR. P and S waves amplifications are represented
materials. Being fine-grained materials usually rich in term with dashed lines. B) best estimate subsurface model. Depth of
of remains, these areas are the best candidates for future deposit is found nearly at the depth of 6 m. An high Vs layer is
excavations. On the contrary, ERT- lines 2 and 3 show the present, nearly 1 m deep, most probably corresponding to the
presence of very high homogenous resistivity volumes that stratigraphic units highligted in figure c). c) An image of the
deposit wall showing closely packed thin stratigraphic units.
© 2017 SEG Page 5127

archaeological material and the almost complete absence of

A relatively shallow anomaly, around 40 cm deep, can be permanent structures. The presence of numerous, thin and
observed in profile no. 3 with regular geometry, which closely packed occupation layers containing archaeological
almost certainly can be associated to an old test pit whose remains generally of tiny dimension that would be
traces are still visible today in the cave-mouth. GPR scans destroyed by invasive investigations makes the contribution
were able to highlight features in accordance with the ERT of integrated geophysical methods significant. Features
result. Figure 2a shows real amplitude view of radar section present on the site such as simple hearts and areas
performed using a 400 MHz antenna along the major ERT associated with the exploitation of large mammals and
anomalies (GPR1 in Figure 1). Grouped reflections may be stone knapping, implicate that the site was periodically
observed. Further, Figure 2b shows a total energy view of occupied by humans which left traces that could be
radar section acquired vertically along the excavated wall detected only using very high resolution geophysical
of the deposit using a 200 MHz antenna. The radar methods such as for example high frequency GPR (> 900
signature corresponding to the fast seismic layer MHZ) to be executed before and during each excavation
highlighted by HVSR can be observed. campaign. Of course this approach has a very short depth
of investigation and is unpractical for many aspects.
The ERT result, as it was expected, did not give any Nevertheless, geophysical methods still retain some value
indication neither about the maximum thickness of the even at lower resolution. The present investigation proved
deposit nor its basal morphology. Therefore, we tested the successful in the description of subsurface morphology and
capability of HVSR to obtain an estimate of the whole possibly in the nature and thickness of sedimentary infill,
thickness of the deposits. Figure 3 show the inversion result delineating potential areas of archaeological interest in a
of the HVSR data acquired at the center of the cave. The completely non-invasive way. This work leads to
outcome both in the data space (i.e. the fit of the curve) and methodological insights about how to improve both
in the parameter space (the shear wave velocity “Vs” efficiency and effectiveness of future archaeological
profile) are shown in figure 3a and 3b respectively. Two campaigns, especially suitable for the Palaeolithic age,
major impedance contrasts can be observed at roughly 2.5 enabling a better and successfully focused management of
and 7 meters respectively. Further the inversion routine the available funds.
highlighted a layer of high Vs values located roughly at 1
meter beneath the surface which seemed to coincide with Being archaeological excavations destructive, it is of
the coarse-grained stratigraphic units visible in figure 3c. outmost importance to correctly document the location of
each survey in order to faithfully reconstruct the subsurface
Of course, since available modelling routines assume a geometries in later times. In this sense, the integration of
subsurface simplified as a stack of infinite parallel layers, geophysical maps with accurate 3D geometrical models
which is an acceptable approximation only when lateral allows to build high quality images capable of enhancing
variations are much greater than the wavelengths into play, historical understanding and providing a powerful
to achieve a good fit of the curve was particularly communication tool to attract a broad public, both
challenging. In particular 3D effects are expected to be not professional and non-professional.
negligible in this complex geometry (Guéguen et al. 2007).
Therefore, inversion was not able to retrieve the correct
amplitude of the HVSR curves. Fortunately, in such a kind
of inversions, amplitude is not as an important parameter as
the frequency position of peaks, which despite of the unmet
modeling assumptions was correctly retrieved. The best
estimate Vs profile is shown in figure 2b. The retrieved
velocity profile for the HVSR measurement point in the
vicinity of the cave entrance (i.e. 5 meters below, at the
base of the excavated portion of sediments) provided a Vs
model in accordance to those retrieved from measurements
performed at the top of the paleolithic deposit. Further, it
can be inferred that the sediments become thinner towards
the inner part of the cave.
Conclusions
Prehistoric sites have been greatly ignored by the wider

geophysical community due to the nature and fragility of
© 2017 SEG Page 5128

EDITED REFERENCES
REFERENCES
Abu Zeid, N., E. Corradini, S. Bignardi, and G. Santarato, 2016, Unusual geophysical techniques in
archaeology — HVSR and induced polarization, a case history: NSAG-2016, 22nd European
Meeting of Environmental and Engineering Geophysics, https://doi.org/10.3997/2214-
4609.201602027.
Abu Zeid, N., E. Corradini, S. Bignardi, V. Nizzo, and G. Santarato, 2017, The passive seismic technique
‘HVSR’ as a reconnaissance tool for mapping paleo-soils: The case of the Pilastri archaeological
site, northern Italy: Archaeological Prospection, https://doi.org/10.1002/arp.1568.
Bignardi, S, A. Mantovani, and N. Abu Zeid, 2016, OpenHVSR: imaging the subsurface 2D/3D elastic
properties through multiple HVSR modeling and inversion: Computers & Geosciences, 93, 103-
113, https://doi.org/10.1016/j.cageo.2016.05.009.
Bolognesi M., M. Obradovic, N. Abu-Zeid, M. Peresani, A. Furini, P. Russo, and G. Santarato, 2015,
Integrazione di rilievi laserscan e fotogrammetrici con metodologie geofisiche applicata ad una
cavità pleistocenica con stratificazione archeologica: Accademia Roveretana degli Agiati, a. 265,
2015, ser. IX, vol. V, B: 21–30.
Di Giulio, G., C. Cornou, M. Ohrnberger, M. Wathelet, and A. Rovelli, 2006, Deriving wavefield
characteristics and shear-velocity profiles from two-dimensional small-aperture arrays analysis of
ambient vibrations in a small-size alluvial basin, Colfiorito, Italy: Bulletin of the Seismological
Society of America, 96, 1915–1933, https://doi.org/10.1785/0120060119.
Guéguen, P., C. Cornou, S. Garambois, and J. Banton, 2007, On the limitation of the H/V spectral ratio
using seismic noise as an exploration tool: Application to the Grenoble valley (France), a small
apex ratio basin: Pure and Applied Geophysics, 164, 115–134, https://doi.org/10.1007/s00024-
006-0151-x.
Nakamura, Y., 1989, A method for dynamic characteristics estimation of subsurface using microtremor
on the ground surface: Quarterly Report of Railway Technical Research Institute, 30, 25–33.
Nakamura, Y., 2000, Clear identification of fundamental idea of Nakamura’s technique and its
applications: Presented at the 12th World Conference on Earthquake Engineering, New Zealand.
Morelli, G., and D. J. LaBrecque, 1996, Advances in ERT inverse modeling: European Journal of
Environmental and Engineering Geophysics, 1, 171–177, http://dx.doi.org/10.4133/JEEG1.3.171.
Obradovic, M., N. Abu Zeid, S. Bignardi, M. Bolognesi, M. Peresani, P. Russo, and G. Santarato, 2015,
High Resolution Geophysical and Topographical Surveys for the Characterisation of Fumane
Cave Prehistoric Site, Italy: Near Surface Geoscience 2015: 21st European Meeting of
Environmental and Engineering Geophysics, https://doi.org/10.3997/2214-4609.201413676.
Peresani, M., 2012, Fifty thousand years of flint knapping and tool shaping across the Mousterian and
Uluzzian sequence of Fumane cave: Quaternary International, 247, 125–150,
https://doi.org/10.1016/j.quaint.2011.02.006.
Peresani, M., M. Cremaschi, F. Ferraro, Ch. Falguères, J. J. Bahain, G. Gruppioni, E. Sibilia, G. Quarta,
L. Calcagnile, and J. M. Dolo, 2008, Age of the final middle palaeolithic and uluzzian levels at
Fumane cave, northern Italy, using 14C, ESR, 234U/230Th and thermoluminescence methods:
Journal of Archaeological Science, 35, 2986–2996, https://doi.org/10.1016/j.jas.2008.06.013.
Wathelet, M., D. Jongmans, M. Ohrnberger, and S. Bonnefoy-Claudet, 2008, Array performances for
ambient vibrations on a shallow structure and consequences over Vs inversion: Journal of
Seismology, 12, 1–19, https://doi.org/10.1007/s10950-007-9067-x.
© 2017 SEG Page 5129

Joint Geophysical Methods for Geoforensic Studies of Leachate Plume Migration from Cemeteries in Benin City,
Nigeria
O.M. ALILE*
Department of Physics, University of Benin, Benin, Edo State, Nigeria. Email: monday.alile@uniben.edu
And
J.O. AIREN
Department of Physics, University of Benin, Benin City, Nigeria. Email:osariere.airen@uniben.edu
Summary conductors to be leachate plumes for Second and Third
The environment on which burial takes place has large Cemeteries, and non-leachate plumes for First Cemetery.
quantities of nutrients such as nitrogen, and chemicals, like Soil samples from shallow holes drilled within the three
lead, that can negatively affect it and cause health problems Cemeteries indicated the presence of contaminants due to
such as cancer and if contaminants come in contact with migration of leachate plumes. The contaminant migration is
drinking water sources. Very Low Frequency- aided by local depressions and poorly consolidated
Electromagnetic (VLF-EM) method has the potential to aid subsurface layer. A time lapse study showed contaminant
the location of these pollutants as it can, non-invasively, migration rates of 41.6 cm/month and 51.7 cm/month in the
very rapidly survey extensive areas where a suspected horizontal directions in the Second and Third Cemeteries
conductive material from burial activity is located. respectively and 19.2 cm/month in the vertical directions
Subsequent targeted conductive anomalies can then be for both cemeteries. Also, the arrival time of migrating
investigated using a well known conventional intrusive plumes in laterite layer under favourable hydrological and
method such as Electrical Resistivity Tomography (ERT). geological conditions was estimated to be 4 years. This
Environmental and criminal geoforensics which involves investigation demonstrates the suitability of Environmental
an integrated suite of geochemical and geophysical and Criminal geoforensics for identification and evaluation
techniques (VLF-EM and ERT) was used to detect and of electrically conductive contaminant plumes, and also to
evaluate contaminant plume migration from three monitor the plume as it travels within the subsurface.
cemeteries in Benin-City, South-South Nigeria, which is

Introduction
aimed at determining the risk of boreholes and soil in the
Agriculture, industry and landfills are commonly believed
study area and also putting a time stamp to leachate plume
to be major anthropogenic sources of environmental
migration. The Very Low Frequency-Electromagnetic
contamination. Little attention has been given to cemeteries
survey pin-pointed locations of vertical and near vertical
as possible pollution sources. Research conducted on the
conductive bodies. The Electrical Resistivity Imaging
survey conducted over the VLF-EM anomalies showed the
© 2017 SEG Page 5130

Joint Geophysical Methods for Geoforensic Studies of Leachate Plume Migration from Cemeteries in Benin City, South-
South Nigeria
latter has been limited to examining pollutants emanating also be found. If the cemetery is on vulnerable soil or if the
from dead bodies. soil reaches its depurative limit, the pollution may reach the
However, cemeteries are not the final resting place to groundwater (Ucisik and Rushbrook, 1998).
bodies only but also to coffins and caskets used for the In 1999, Young et al collected the information available at
interment of remains. Studies conducted found that the that time about the potential threats that burial sites may
highest contamination arising from cemeteries originated pose to the environment. Different reports on this subject
from minerals that are released by burial loads (Borstel and confirm that decomposition forms a saline contamination
Niquette, 2000). The minerals that are used in Coffin- plume when geological, hydrogeological and climatic
making may corrode or degrade releasing harmful toxic factors are not taken into account when choosing the
substances (Spongberg and Becks, 2000b). These may be locations of new burial grounds.
transported from the graves through seepage and diffuse Groundwater applications of near-surface geophysics
into surrounding soils. From there they may leach into include mapping the depth and thickness of aquifers,
groundwater and become a potential health risk to the mapping aquitards or confining units, locating preferential
residents in areas surrounding the cemetery (Jonker and fluid migration paths such as fracture and fault zones and
Olivier, 2012; Engelbrecht, 2010; Dent and Knight, 1998; mapping contamination to the groundwater such as that
Kim et al, 2008; Williams et al, 2009; Canninga and from saltwater intrusion (Benson et al, 1997; Kalik and
Szmigina, 2010) Most existing cemeteries were sited Kaya, 2001).
without thinking about potential risks to the local This research work presents an indirect method and a cost
environment or community (WHO Nancy Project Report- effective way of accessing the impact of cemeteries on the
TARGET 23, 2000). soil and groundwater regime without resorting to drilling
In 1998, the World Health Organization (WHO) published which is usually too expensive for the purpose.
a short review of soil and groundwater contamination by Methods
cemeteries with the aim of evaluating its impacts on the The approach used in this study is similar to that adopted
environment and public health. The main conclusion is that by Alile et al (2012) where they carried out Groundwater
buried corpses have different microbial organisms, and the Contaminant study in Ota, Southwestern Nigeria, using
materials used in funeral practices may be sources of subsurface Geophysical Investigation and Physiochemical-
chemical compounds and heavy metals. Nuclear waste Microbial Analysis. An integrated suite of both Very Low
from medical treatments or devices and prostheses may Frequency Electromagnetic and Electrical Resistivity
© 2017 SEG Page 5131

South Nigeria
Tomography was used as emphasized by Ruffell (2013). East-West direction and spaced equally at 2 m. The
Ruffell (2013) considered the importance of traditional electromagnetic response gave real and imaginary values
geological data, such as maps, cross-sections and measured and were read and recorded at 5 m interval along each line.
sections in both criminal and environmental forensic The data were analyzed by preparing curves of filtered real
investigations. and filtered imaginary values with station distances in
This research work commenced after permission from the Matlab Graphical User Interface (MGUI), and contour
Government of Edo State, Nigeria through the Ministry of maps of the filtered real values drawn in surfer 11.0. The
Environment, Benin City was sought, so as to have free VLF-EM Pseudosections were constructed using KHFFILT
access to the Cemeteries. A reconnaissance survey was software. The combined curves and 2D pseudosections
carried out in the three Cemeteries. Geochemical survey were interpreted to locate near surface conductive
was conducted to monitor differences of subsurface anomalies suspected to be the leachate plumes detected by
activities within and outside the cemeteries. Soil and water the geochemical analysis. The VLF-EM anomalous points
samples were taken and analyzed at Quality Analytical were picked where the filtered real curves showed positive
Laboratory Services Ltd., located in Benin City. The pH bulge and at shallow depths in the corresponding VLF-EM
and EC were obtained using pH and EC meters Pseudosections.
respectively. Calcium, magnesium, chloride and alkalinity The array type used for the Electrical Resistivity
were analyzed using the titration method. Nitrate, sulphate Tomography (ERT) investigation is dipole-dipole
and phosphate were analyzed with a UV spectrophotometer configuration. This array type is most sensitive to
and heavy metals with an Atomic Absorption resistivity changes between the electrodes in each dipole
Spectrophotometer (AAS). pair. The dipole-dipole array is very sensitive to horizontal
Around each cemetery, three holes were drilled, two within changes in resistivity, but relatively insensitive to vertical
and one outside (as control) with hand auger to a depth of 5 changes in the resistivity (Loke, 2014).
ft (1.524 m). Water samples were collected at boreholes The ERT survey was executed at two different times. The
close to and far away from each cemetery in a triangular first was conducted in the month of August, 2014 while the
pattern. second exactly one (1) year later. The first ERT survey was
The VLF-EM data were collected using Abem Wadi in used to determine the nature, depths, vertical and horizontal
arrays of 16 parallel lines each of length 100 m in First and extent of the conductors located by the VLF-EM Survey.
Second Cemeteries and 60m in Third Cemetery, oriented in
© 2017 SEG Page 5132

South Nigeria
The apparent resistivity data from first ERT were inverted Criminal Geoforensics was successfully carried out in the
using Res2dinv software to obtain 2D model of the true three cemeteries in Benin City. This research work carried
subsurface resistivities. out using Joint Geophysical Methods and hydro
The 2D geoelectrical images were used to interpret the physicochemical analysis showed that the arrival time of
VLF-EM anomalies. The positions of the EM anomalies the plumes to the next layer (just below the laterite layer) in
were plotted into the prepared ERT survey location base the second and third cemeteries is estimated to be 4 years.
map, and the corresponding positions on the ERT survey This layer is at shallow depth and is not where boreholes
lines were noted, and used to pick out the resistivity are constructed.
anomalies on the geoelectrical images. Surface and subsurface soil investigation in the burial
The second ERT survey was used to check for environment showed that it is purely laterite, which is
displacement of the plumes in vertical and horizontal impervious to fluid flow. Generally, many depressions
directions, identified in the first ERT survey. This was were identified within the study area, although migration
conducted on the same profile of the first ERT survey using rate is low because the rate of flow is controlled mainly by
the same equipment for data collection and processing subsurface geology. But in an area where there is surface
software. The vertical and horizontal migrations, migration well and the top soil is mainly sand, the rate of migration is
rates and arrival times (predicted) of plumes at the sandy faster and such area is prone to high risk of pollution.
layer just below the surface layer were then computed. The geochemical analysis of soil samples taken at shallow
depths compared to control show high chloride content,
Conclusion high electrical conductivity and high Total Dissolved Solid
Many subdisciplines within the Earth Sciences make use of (TDS) from inorganic chemicals in the plumes which is in
either geophysical instrument to investigate the subsurface agreement with low resistivity (less than 120 Ωm)
environment or use analytical methods to determine the anomalies in the ERT survey.
provenance of geological materials. Under favourable hydrological and geological conditions,
These same instrument and analytical methods can be used the plumes delineated from the Electrical Resistivity
either directly or adapted to suit the acquisition of data that Images will slowly migrate into the aquifers.
pertain to a wide range of forensic science investigations. The result of this study also point out the need for
With recent concerns about the environmental impact of environmental education and proper management/location
decomposing corpses and burial items, Environmental and of cemeteries by Nigeria Government.
© 2017 SEG Page 5133

EDITED REFERENCES
REFERENCES
REFERENCES
Alile, O. M., S. Oranusi, O. Adetolano, and J. O. Airen, 2012, Subsurface geophysical investigation and
physiochemical/microbial analysis of groundwater contaminant in Ota, Southwestern Nigeria:
Geosciences, 2, 179–184, http://doi.org/10.5923/j.geo.20120206.05.
Hansen, J. D., and J. K. Pringle, 2013, Comparison of magnetic, electrical and ground penetrating radar
surveys to detect buried forensic objects in semi-urban and domestic patio environments:
Geological Society of London, Special Publications 13, 384, 229–251,
http://doi.org/10.1144/SP384.13.
Larizza, M., and S. L. Forbes, 2013, Detection of fatty acids in then lateral extent of the cadaver
decomposition island: Geological Society of London, Special Publications 11, 384, 209–219,
http://doi.org/10.1144/SP384.11.
Loke, M. H., 2004, Tutorial: 2-D and 3-D electrical imaging surveys: Geotomo Software, Inc.
Reynolds, J. M., 1998, An introduction to applied and environmental geophysics, 2nd ed.: John Wiley
and Sons Ltd., 423.
Ruffel, A., 2013, Solid and drift geology in forensic investigations: Geological Society of London,
Special Publications 15, 384, 163–172, http://doi.org/10.1144/SP384.15.
© 2017 SEG Page 5134

GPR Polarization Effects of Buried Bison Bone
Schneider, B.B.*, Department of Geology, University of Kansas
Tsoflias, G.P., Department of Geology, University of Kansas
Summary signal from the transmitting antenna and scattering

properties of the subsurface targets (Radzevicius and
Recent advances in ground-penetrating radar (GPR) Daniels, 2000). GPR polarization effects created by dipping
research highlight the importance of polarization effects planes and metal cylinders are described in detail by
when designing surveys. In particular, polarization Roberts and Daniels (1996). Radzevicius and Daniels
properties can be used to determine the size, shape, and (2000) furthered this research by comparing polarization
orientation of buried objects. This research examines GPR effects of metal cylinders versus plastic cylinders with
polarization effects when imaging modern bison bone, varying diameters. These results are supported by van der
which can serve as a proxy for the detection of prehistoric Kruk et al. (2010) who simulated metallic and dielectric
animal bone at archaeological sites. Controlled sandbox pipes with diameters of 0.2 m embedded in a homogenous
experiments show that modern bison bone is imaged sand. The dielectric pipes were evaluated at both lower and
successfully using GPR and the best results are obtained higher relative permittivity values as compared to the
when the dipole antennas are oriented perpendicular to the surrounding sand. The results of these previous
long axis of the bone. investigations are summarized in Table 1. This paper
presents results of polarization effects of animal bone.
Introduction Preliminary results from this investigation show that
modern bison bone, which has a higher relative permittivity
Ground-penetrating radar is a popular subsurface imaging than the surrounding sand medium and a diameter
method for archaeological exploration. Data acquisition is equivalent to approximately one wavelength, is best imaged
quick and non-invasive, which is an important when the antennas are oriented perpendicular to the long
consideration when trying to identify archaeological sites axis of the bone.
that contain sensitive cultural remains. Additionally, users
can collect high-resolution 3D grids prior to excavation or
in lieu of excavation entirely, which is more cost-effective Subsurface Target Description
than shovel testing or excavating test grids. Previous Horizontal interface Best imaged with the T and R
research has successfully applied GPR to detect historic antennas oriented parallel to
and prehistoric archaeological features such as buried each other.
foundations, graves and prehistoric earthworks (Schneider Dipping interface Best imaged when the strike
et al., 2016; Bigman, 2014; Urban et al., 2014). of the layer is parallel to the T
and R antenna orientation.
Animal bone is frequently encountered at prehistorical Metallic cylinders Best imaged when the T and R
archaeological sites, usually contained within cultural antennas are orientated
features or even in dense bone-beds (Davis, 1978). Animal parallel to the long axis of the
bone has been used by archaeologists to determine past pipe.
climates (Leyden et al., 2006), provide insights into socio Small diameter (<1/10λ), Best imaged with the T and R
and economic histories of identified sites (Uerpmann, dielectric pipes with antennas oriented
1973), and provide ages for sites using radiocarbon dating permittivity higher than perpendicular to the long axis
(Hoard et al., 2004; Aitken, 1990). However, despite the surrounding medium of the pipe.
important information that animal bone provides, little Dielectric pipes (diameter Varies depending on the
research has been directed towards detecting buried animal >1/10λ) with permittivity diameter of the pipe and the
bone using non-invasive geophysical methods. Our higher than surrounding central frequency
previous research revealed that animal bone is a good medium
dielectric target for GPR investigations (Schneider and Dielectric pipes with Best imaged with the T and R
Tsoflias, 2016). This research presents new insight into permittivity lower than antennas oriented
GPR’s ability to detect buried animal bone by investigating surrounding medium perpendicular to the long axis
GPR polarization effects on buried bison bone. of the pipe.
Many commercially available GPR systems employ dipole Table 1: A summary of the polarization results for various
subsurface targets obtained from van der Kruk et al. (2010),
antennas which generate linearly polarized electromagnetic
Radzevicious and Daniels (2000), Roberts and Daniels (1996). T
waves. The polarization of the signal measured by the and R indicate transmitting and receiving antennas respectively.
receiver antenna is a function of the polarization of the
© 2017 SEG Page 5135

GPR Polarization Effects of Animal Bone
Methodology orientations, as well as cross-polarized antenna pairs

(Figure 2). Polarization data was collected sequentially,
GPR data was collected over a wooden sandbox that with all six lines completed in approximately 10 minutes, in
measured 2 m long by 1 m wide and 1 m tall. The box was order to minimize instrument drift and changes in
filled halfway with dry, fine play sand and then a bison environmental conditions. Minimal processing was
bone humerus was placed atop this surface at a depth of 0.5 completed in an effort to preserve signal amplitudes for
m from the top of the box. In addition, an aluminum pipe comparison. Processing steps included time-zero
was placed atop this surface at a depth of 0.5 m from the adjustment, trimming the time window, and applying a
top of the box. The bison bone was approximately 45 cm bandpass filter from 500 MHz to 2 GHz for each dataset.
long with a 5.4 cm diameter and was placed 0.6 m from the
south end of the box. The aluminum pipe was 45 cm long
with a 1.8 cm diameter and was placed 0.6 m from the
north end of the box (Figure 1). The remainder of the box
was then filled with more fine sand and a 2 mm thick
Plexiglass sheet was placed atop to ensure a smooth surface
for GPR data collection. The box sits over smooth concrete
floor, which provides an additional reflector for
comparison.
Figure 2: The six antenna orientations used for this experiment.

BR:broadside, EF:endfire, and XPOL:cross-polarized antenna
orientation. Antennas were oriented parallel and perpendicular to
the acquisition line. Transmitters are indicated by white boxes,
receivers by black boxes.
Results
The results of the co-pole antenna orientations are shown in

Figures 3 and 4. Figure 3 shows the GPR inline for each
orientation in the center of box. The location of the buried
bone versus buried pipe is marked and is seen at
approximately 7 ns. For reference, the concrete floor
reflection is located 1 m below the top at approximately 12
Figure 1: Top down view of the experiment design (top) and ns. As shown in previous studies, the pipe is clearly imaged
corresponding photo (bottom). best when the antennas are oriented parallel to the long axis
of the pipe in both the broadside and endfire orientation.
GPR data was collected using a Sensors and Software Pulse The bone specimen, however, is best imaged when the
EKKO PRO system with 1 GHz frequency antennas. The antennas are aligned perpendicular to the long axis of the
sampling interval was set at 0.1 ns with a total time window bone for both broadside and endfire orientations.
of 25 ns and a trace spacing of 0.02 m. A total of six
different antenna orientations were deployed, including co- Figure 4 compares the corresponding traces over the center
polarized antenna pairs in both the broadside and endfire of the bone at the four co-pole orientations. For each
© 2017 SEG Page 5136

Figure 3: GPR inlines for the a) broadside antennas perpendicular to the long axis of the pipe and bone, b) broadside antennas parallel to the long
axis of the pipe and bone, c) endfire antennas perpendicular to the long axis of the pipe and bone, and d) endfire antennas parallel to the long axis
of the pipe and bone.
corresponding trace, the RMS amplitude was calculated

over one period of the bone and pipe reflection. The RMS
amplitude of the aluminum pipe with the antennas endfire
and parallel to the long axis is 65% greater than the
amplitude of the pipe with the antennas perpendicular to
the long axis.. The RMS amplitude of the aluminum pipe
with the antennas broadside and parallel to the long axis is
54% greater than the amplitude of the pipe with the
antennas perpendicular to the long axis. The RMS
amplitude of the bison bone with the antennas broadside
and perpendicular to the long axis of the bone is 46%
greater than the amplitude of the bone reflection with the
antennas parallel to the long axis. The RMS amplitude of
the bison bone with the antennas endfire and perpendicular
to the long axis of the bone is 36% greater than the
amplitude of the bone with the antennas parallel to the long
axis.
The overall amplitude scale of the entire trace was twice as

high when collecting in broadside orientation versus
endfire orientation, but the amplitude values over the bone
target were similar. There was less than 10% change in the
amplitudes between broadside and endfire orientations
when comparing amplitudes over the bone. RMS
Figure 4: Corresponding traces over the bone target for the four amplitudes were 9% higher when collected broadside and
co-pole antenna orientations. Broadside co-pole orientations are perpendicular to the bone long axis and 7% higher when
on top and endfire co-pole orientations are on the bottom. The collected endfire and parallel to the bone long axis. It can
bone reflection is at approximately 7 ns and the concrete floor at
12 ns.
© 2017 SEG Page 5137

also be seen in Figure 4, as well as Figure 3b and 3d, that metallic and dielectric pipes are best imaged based on their
there is a longer reverberation below the bone target when radius and relative permittivity ratio with their surrounding
the antennas are oriented parallel to the bone long axis than medium. This investigation expands on that research by
when they are perpendicular to the bone long axis, investigating how multi-polarization GPR data can be used
regardless of orientation. to better detect buried animal bone at archaeological sites.
Our results show that imaging animal bone exhibits strong
Figure 5 shows the GPR lines for the two cross-pole polarization effects. For this investigation, the animal bone
antenna orientations. As expected, the amplitude range was was best imaged when the antennas were perpendicular to
greatly reduced as compared to the co-pole antenna the long axis of the bone target, regardless of broadside
orientations. The diffraction from the buried bone target is versus endfire orientation. Endfire orientations are not used
almost undetectable next to the diffraction from the buried as frequently in GPR studies because broadside orientations
pipe target, but appears slightly stronger when the offer better coupling when collecting data. It is still
transmitter is oriented parallel to the orientation of the bone important to investigate GPR data in this orientation
long axis and the receiver is oriented perpendicular to the though, because some sites may require an endfire
bone long axis. The same RMS amplitude analysis was orientation acquisition. This could be due to space
completed over both subsurface targets. The aluminum pipe limitations when setting up the grid, or if single borehole
was consistent across both antenna orientations, with only a GPR data is being collected.
5% increase in amplitude when the transmitter was oriented
perpendicular to the long axis and the receiver was oriented Additional observations have shown that there appears to
parallel to the long axis. There is also a phase shift noted be some dependence on the bone diameter, which is
between the two antenna orientations over the pipe. The expected from the results of previous investigations. Here
RMS amplitude analysis over the bone target showed a we only present one single experiment to show that when
58% increase in amplitude when imaged using the imaging bone polarization effects should be considered.
transmitter parallel to the bone long axis and the receiver Further investigations should analyze polarization effects
antenna oriented perpendicular to the bone long axis. with changes in animal bone size and type, as well as how
increased weathering affects polarization effects of buried
prehistoric animal bone.
Acknowledgements
We would like to acknowledge the University of Kansas

Archaeology division of the Biodiversity Institute and
Natural History Museum for their donation of samples for
this research. The GPR instrumentation acquisition was
funded by the National Science Foundation EAR/IF-
0345445.
Figure 5: GPR inlines for the a) cross polarized antennas with

transmitter parallel to the long axis of the pipe and bone and b) cross
polarized antennas with transmitter perpendicular to the long axis of
the bone and pipe.
Conclusions
Multi-polarization GPR data is an important design tool

that can be used to identify the size, shape and orientation
of buried archaeological features and artifacts. Previous
research has investigated how cylindrical targets like
© 2017 SEG Page 5138

EDITED REFERENCES
REFERENCES
Aitken, M. J., 1990, Science-Based Dating in Archaeology: Routledge, New York.
Bigman, D.P., 2014, Mapping social relationships: geophysical survey of a nineteenth-century American
slave cemetery: Archaeological and Anthropological Sciences, 6, 17–30,
http://dx.doi.org/10.1007/s12520-013-0119-6.
Davis, L. B., 1978, Bison Procurement and Utilization: A Symposium: Plains Anthropologist, 23, 362.
Hoard, R. J., W. E. Banks, R. D. Mandel, M. Finnegan, and J. E. Epperson, 2004, A middle archaic burial
from east central Kansas: American Antiquity, 69, 717–739, http://dx.doi.org/10.2307/4128445.
Leyden, J. J., L. I. Wassenaar, K. A. Hobson, and E. G. Walker, 2006, Stable hydrogen isotopes of bison
bone collagen as a proxy for Holocene climate on the Northern Great Plains: Palaeogeography,
Palaeoclimatology, Palaeoecology, 239, 87–99, http://dx.doi.org/10.1016/j.palaeo.2006.01.009.
Radzevicius, S. J. and J. J. Daniels, 2000, Ground penetrating radar polarization and scattering from
cylinders: Journal of Applied Geophysics, 45, 111–125, http://dx.doi.org/10.1016/S0926-
9851(00)00023-9.
Roberts, R. L. and J. J. Daniels, 1996, Analysis of GPR polarization phenomena: Journal of
Schneider, B. B., R. M. Mandel, G. Tsoflias, S. L. De Vore, and M. Lynott, 2016, Combining ER and
GPR surveys for evidence of prehistoric landscape construction: case study at Mound City, Ohio,
USA: Journal of Applied Geophysics, 129, 178–186,
http://dx.doi.org/10.1016/j.jappgeo.2016.04.002.
Schneider, B. B., and G. Tsoflias, 2016, Modelling permittivity for the application of GPR for detecting
buried prehistoric animal bones: Geological Society of America Abstracts with Programs, 48,
http://dx.doi.org/10.1130/abs/2016AM-282979.
Uerpmann, H., 1973, Animal bone finds and economic archaeology: a critical study of ‘Osteo-
Archaeological’ method: World Archaeology, 4, 307–322.
Urban, T. M., J. F. Leon, S. W. Manning, and K. D. Fisher, 2014, High resolution GPR mapping of Late
Bronze Age architecture at Kalavasos-Ayios Dhimitrios, Cyprus: Journal of Applied Geophysics,
107, 129–136, http://dx.doi.org/10.1016/j.jappgeo.2014.05.020.
van der Kruk, J., R. Streich, and M. Grasmueck, 2010, Toward true-amplitude vector migration of GPR
data using exact radiation patterns, in R. D. Miller, J. H. Bradford, K. Holliger, K., eds.,
Advances in near-surface seismology and ground-penetrating radar: Society of Exploration
Geophysicists, 97–1, https://doi.org/10.1190/1.9781560802259.
© 2017 SEG Page 5139

Sparse Blind Deconvolution of common-offset GPR data
Sajad Jazayeri1*, Alaeddin Ebrahimi2, Sarah Kruse1, 1. School of Geosciences, University of South Florida, Tampa,
FL, USA. 2. Earth Science Department, Sultan Qaboos University, Oman.
Summary a FWI formulation to the most widely used mode of GPR

data acquisition, namely common-offset GPR, to locate and
Ground penetrating radar (GPR) is widely used for shallow resolve isolated targets such as buried utilities and UXOs.
(cms to tens of meters) subsurface imaging. Full-waveform
inversion (FWI) of GPR data has enabled researchers to An essential yet challenging factor for the FWI problem is
increase the subsurface image resolution. The FWI the estimation of the source wavelet before or during the
technique has been applied primarily to off-ground GPR and FWI procedure. For GPR source wavelet estimation, the
crosshole GPR data, due to complexity of surface-based on- deconvolution method introduced by Ernst et al. 2007 and
ground data. A major difficulty with common-offset surface Klotzsche et al. 2010 has been used widely. This method
GPR data is the source wavelet estimation, particularly due requires a good initial synthetic model of the subsurface
to presence of the air wave, ground wave and noise. Existing electrical permittivity and conductivity structures. Errors in
deconvolution methods for estimation of the source wavelet the initial model of the subsurface can propagate into the
require preliminary knowledge of the subsurface. Sparse source wavelet estimation, which in turn can potentially
blind deconvolution (SBD) methods permit estimation of the generate significant errors in the FWI results.
source wavelet without an initial synthetic model of soil
structure. We investigate the performance of an SBD method A method that does not require such a good initial model of
to estimate the source wavelet of common-offset GPR data the subsurface is clearly desirable. Blind deconvolution
The effects of including/excluding the air/ground wave are techniques that do not require any initial soil structure model
studied on both synthetic and simple field test data involving offer a possible path forward (Wang et al., 2015).
a single buried pipe. For the 2D synthetic model, SBD
extracts the wavelets. For the real field data, the estimated Gholami and Sacchi 2012 developed a SBD method for the
wavelets are compared to those derived from the estimation of the source wavelet for seismic data. Ebrahimi
deconvolution method. Ongoing research will examine the et al., 2017 applied this technique to derive a deconvolved
relative quality of FWI inversion results based on wavelets GPR section, in order to to increase the vertical resolution of
estimated with SBD.. the GPR profile. In this paper, we examine the efficiency of
Gholami and Sacchi’s (2012) SBD method to estimate the
Introduction source wavelet of common-offset GPR data, and compare
the results against the well-established deconvolution
GPR provides subsurface information using high-frequency method of Ernst et al (2007) and Klotzsche (2010). The SBD
electromagnetic waves. Due to its efficiency and non- estimation process is completely automated and the
invasiveness, GPR is a highly popular exploratory method generalized cross validation (GCV) technique is used for the
with wide range of applications in geophysics, calculation of the regularization parameter. Since on-surface
hydrogeology, archaeology and engineering (e.g. Daniels GPR data (specifically common-offset data) are highly
2004, Klotzsche, et al. 2010, and Benedetto 2015). Because impacted by the effects of the air wave, ground wave, and
many of the principles and problems of GPR data acquisition surface coupling, we investigate two schemes for data
and processing are analogous to seismic surveying, many selection. The source wavelets are estimated both with and
GPR data processing techniques are adapted from seismics without including the air/ground wave to study the influence
(e.g. Fisher et al. 1992 and 1996, Annan, 1992, Young et al., of the first received portion of energy on the trace. Synthetic
1995). and real data are tested. The results for the real data are
compared to the deconvolution method results.
Tarantola, 1984 is one of the first studies in exploration
geophysics that used the full-waveform seismic data to Method
increase resolution of subsurface images. Since then, full
waveform inverseion (FWI) has been the subject of ongoing In the common-offset data gathering process, transmitter and
development. Virieux and Operto, 2009 provides a receivers move together at a constant distance from each
comprehensive study on the history of FWI technique other. The simplest form of common-offset data can be
development for seismic data. Researchers now utilize this collected using shielded commercial antennas, while the
technique with GPR data. Most efforts reported to date have transmitter and receiver are fixed in a box next to each other.
been applied to (a) off-ground GPR aimed at soil properties Following Gholami and Sacchi (2012) the GPR trace 𝑬(𝒕)
(e.g. Lambot et al., 2004; Andre and Lambot, 2015) or recorded at the receiver is the earth response to the
concrete (e.g. Kalogeropoulos et al., 2011), or (b) crosshole transmitted source wavelet 𝒘(𝒕) emitted from the
GPR data (e.g. Klotzsche et al., 2010 and 2013; Busch et al., transmitter antenna, plus noise. 𝑬(𝒕) can be calculated by
2012 and 2013; Gueting et al., 2017). More recently, Busch convolving 𝒘(𝒕) and subsurface reflection series 𝑹(𝒕) and
et al., 2014 demonstrated that FWI applied to common-mid- adding a noise term 𝒆(𝒕).
point (CMP) GPR data can improve soil conductivity and
permittivity models. Jazayeri et al. (2016a and 2016b) apply 𝑬(𝒕) = 𝒘(𝒕) ∗ 𝑹(𝒕) + 𝒆(𝒕) (1)
© 2017 SEG Page 5140

where * denotes the convolution operator. Both the ||𝑯𝒅𝒌 −𝑬||𝟐

𝑮𝑪𝑽(𝒌) = [𝑵−𝑪.𝒏𝒏𝒛(𝒅𝒌𝟐)]𝟐 (6)
reflectivity series and the source wavelet are unknown for
the real data. The best source wavelet is defined as that
which minimizes the following second norm statement in the where H is the square Toeplitz matrix of the wavelet (𝑯 =
frequency domain (2): 𝑭−𝟏 𝑾𝑭), dk is the solution of equation (4) for the kth
iteration, N is the number of data points in the trace, nnz is

𝒂𝒓𝒈𝒎𝒊𝒏 ||𝑭𝑬 − 𝒘||𝟐𝟐 + 𝝀||𝑫𝟐 𝒘||𝟐𝟐 (2) the number of non-zero elements of a vector and C is the
stabilizing parameter which is used to restrict the maximum
allowed iterations.
where 𝑭 is the Fourier transform matrix, 𝑫𝟐 is the second
derivative operator, and 𝝀 is the regularization parameter. An
initial source wavelet can be estimated via: Synthetic model
𝑻 This SBD method is evaluated by creating an extremely

̂ = 𝑭𝑻 (𝑰 + 𝝀𝑫𝟐 + 𝑫𝟐 )−𝟏 𝑭𝑬
𝒘 (3)
simple 2D synthetic model of a water-filled pipe in
homogeneous soil (Figure 1). The 4th Gaussian derivative is
where 𝒘 ̂ is the source wavelet in the Fourier domain and 𝑰 used as the source wavelet with the antenna center frequency
is the identity matrix. Using this initially estimated wavelet, is set to 800 MHz. Figure 1 shows the model geometry and
the sparse reflectivity series is obtained from: the synthetic GPR profile. The middle trace of the profile, at
0.5 meter, is selected for SBD and shown on Figure 2.
𝒂𝒓𝒈𝒎𝒊𝒏 ||𝑹||𝟏 𝒘𝒉𝒊𝒍𝒆 ||𝑬 − 𝒘𝒍𝟏 ∗ 𝑹||𝟐𝟐 ≤ 𝜸 (4)
and γ = error bound. The estimation goal from the first

statement in equation 4 is to find the sparsest reflection series
that produces the most similar synthetic data to the real data,
in an L1 norm sense.
Continuing to follow Gholami and Sacchi (2012), the

estimated reflection series should be used to update the
source wavelet in a least squares approach. This process can
be continued until convergence. Equation 4 is solved using
the Bregman iteration provided in the following algorithm:
Figure 1: Left: Geometry of the 2D synthetic model, water-
filled pipe buried in soil. Right: GPR Profile.
Figure 2: Synthetic GPR trace over the center of the pipe

(Figure 1), selected for the SBD process.
The SBD method is applied on the synthetic trace in Figure

2, (1) using the whole 16 ns synthetic trace, and (2)
excluding the air/ground wave by only considering energy
Figure 1: Split Bregman algorithm (from Gholami and arriving after 2 ns. In both cases the SBD method recovers
Sacchi 2012) the source wavelet (Figure 3). The plot of the estimated
wavelet for the air wave-excluded case cannot be
where 𝑾 is the square diagonal matrix of w and the prox distinguished from the wavelet estimated from the whole
(proximity) operator is defined as: trace, so just one of the estimated wavelets is shown on
Figure 3 in blue.
𝒑𝒓𝒐𝒙𝟏𝝉 (𝑬) = 𝒔𝒊𝒈𝒏(𝑬)𝐦𝐚𝐱(|𝑬| − 𝝉, 𝟎) (5)
where τ = the stabilizing proximity constant. The GCV

score is used to find the optimum number of iterations
(Gholami and Sacchi 2012)
© 2017 SEG Page 5141

Klotzsche et al. 2010) source wavelet, excluding air/ground

wave.
Figure 3: The original synthetic and estimated source

wavelets.
Real Data
Figure 5: Trace 40 from the field test over the buried pipe
A water-filled 8.2 cm PVC pipe was buried in sand in the shown in Figure 4..
Geopark of the University of South Florida campus in
Tampa, FL, USA, at depth of 35 cm below the surface. The
top 1-meter soil is principally well-sorted loose sand.
Common-offset GPR data were collected using the MALA
ProEx system with shielded 800 MHz antennas. Simple
filtering steps are applied including dewow and zero-time
correction, but no gain is applied (Figures 4 and 5). Jazayeri
et al. 2016a performed FWI on this dataset and could
estimate the pipe diameter with <8% error and pipe depth
with < 1% error. They used the deconvolution method (Ernst
et al., 2007; Klotzsche et al., 2010) to estimate the source
wavelet, excluding the air/ground wave. We perform the
Gholami and Sacchi (2012) SBD technique to compare the
SBD-derived source wavelet with the “deconvolution”
wavelet.
Figure 6: For center trace from field data (Figures 4 and 5).
Top: Source wavelets derived from SBD. Blue line =
including air/ground wave; red dashed line excludes
air/ground wave. Bottom: source wavelet derived from
deconvolution (Ernst et al., 2007; Klotzsche et al., 2010)
excludes air/ground wave.
The overall shapes of the estimated source-wavelets from the

field data are similar, but with discernable differences. The
deconvolution-derived source wavelet is more asymmetric,
with a more pronounced leading negative excursion
followed by a higher amplitude peak. Somewhat
surprisingly, the addition of the air wave in the SBD yields
a source wavelet more similar to that from the deconvolution
method, which excluded the airwave.
Figure 4: The GPR profile over the water-filled pipe buried
at 35 cm depth in clean sands. The trace interval is 8.5 mm; Jazayeri et al. 2016a showed that the deconvolved wavelet
the profile length is 1 m. presented in Figure 6 bottom leads the FWI process to < 8%
error in estimating inner and outer PVC pipes diameters. To
Trace number 40 in Figure 4, is selected for the SBD process equivalently assess the SBD method, full waveform
and is shown on Figure 5. The air/ground wave is recorded inversions are needed with the SBD-derived wavelet to
before 5 ns. The reflected energy from the top of the pipe is determine the quality of pipe diameter and position
recorded after 6 ns. The source wavelet is estimated (1) using estimations. These inversions are the subject of ongoing
the whole 20-ns trace and (2) using only energy after 6 ns, research.
effectively excluding the air/ground. Figure 6 top shows the
SBD-estimated source wavelets for the whole trace (blue) Conclusions
and for the air-wave excluded wave (dashed red); Figure 6
bottom shows the deconvolution-derived (Ernst et al., 2007;
© 2017 SEG Page 5142

The sparse blind deconvolution (SBD) technique (Gholami

and Sacchi 2012) is a powerful tool that can be used for the
source wavelet estimation of ground penetrating radar data.
Unlike the conventional deconvolution methods that require
an initial synthetic subsurface model, the SBD method needs
no prior information on the subsurface. The SBD method

thus offers a possible advantage over the widely used
deconvolution method of Ernst et al. (2007) and Klotzsche
et al. (2010).
The SBD method is tested on 2D synthetic and real common-

offset GPR data collected over cylindrical targets. It is
shown that the overall shape of the wavelet can be retrieved
using the SBD process. Further investigation is needed to
evaluate the applicability of the SBD derived source
wavelets on FWI of common-offset GPR data and to study
its performance on more complicated and noisier GPR data.
© 2017 SEG Page 5143

EDITED REFERENCES
REFERENCES
Andre, F., and S. Lambot, 2015, Determination of medium electrical properties through full-wave
modelling of frequency domain reflectometry data: Geophysical Research Abstracts 17,
EGU2015-11068.
Annan, A. P., 1992, Ground penetrating radar workshop notes: Sensors and Software Inc.
Benedetto, A., and L. Pajewski, 2015, Civil engineering applications of ground penetrating radar:
Springer, Springer Transactions in Civil and Environmental Engineering.
Busch, S., and J. van der Kruk, 2014, Improved characterization of fine-texture soils using on-ground
GPR full-waveform inversion: IEEE Transactions on Geoscience and Remote Sensing, 52, 3947–
3958, http://dx.doi.org/10.1109/TGRS.2013.2278297.
Busch, S., J. van der Kruk, J. Bikowski, and H. Vereecken, 2012, Quantitative conductivity and
permittivity estimation using full-waveform inversion of on-ground GPR data: Geophysics, 77,
no. 6, H79–H91, http://dx.doi.org/10.1190/GEO2012-0045.1.
Busch, S., L. Weihermller, J. A. Huisman, C. M. Steelman, A. L. Endres, H. Vereecken, and J. van der
Kruk, 2013, Coupled hydrogeophysical inversion of time-lapse surface GPR data to estimate
hydraulic properties of a layered sub-surface: Water Resources Research, 49, 8480–8494,
http://dx.doi.org/10.1002/2013WR013992.
Daniels, D. J., 2004, Ground penetrating radar, 2nd ed.: The Institution of Electrical Engineering, IEE
Radar, Sonar and Navigation Series 15.
Ebrahimi, A., A. Gholami, and M. Nabi-Bihdendi, 2017, Sparsity-based GPR blind deconvolution and
wavelet estimation: The Journal of Indian Geophysical Union, 21, 7–12.
Ernst, J. R., A. G. Green, H. Maurer, and K. Holliger, 2007, Application of a new 2D time domain full-
waveform inversion scheme to crosshole radar data: Geophysics, 72, no. 5, J53–J64,
http://dx.doi.org/10.1190/1.2761848.
Fisher, E., G. A. McMechan, and A. P. Annan, 1992, Acquisition and processing of wide-aperture ground
penetrating radar data: Geophysics, 57, 495–504, http://dx.doi.org/10.1190/1.1443265.
Fisher, S., R. R. Stewart, and H. M. Jol, 1996, Ground penetrating radar (GPR) data enhancement using
seismic techniques: Journal of Environmental and Engineering Geophysics, 1, 89–96,
http://dx.doi.org/10.4133/JEEG1.2.89.
Gholami, A., and M. D. Sacchi, 2012, A fast and automatic sparse deconvolution in the presence of
outliers: IEEE Transactions on Geoscience and Remote Sensing, 50, 4105–4116,
http://dx.doi.org/10.1109/TGRS.2012.2189777.
Gueting, N., T. Vienken, A. Klotzsche, J. van der Kruk, J. Vanderborght, J. Caers, H. Vereecken, and A.
Englert, 2017, High resolution aquifer characterization using crosshole GPR full-waveform
tomography: Comparison with direct-push and tracer test data: Water Resources Research, 53,
49–72, http://dx.doi.org/10.1002/2016WR019498.
Jazayeri, S., A. Klotzsche, and S. Kruse, 2016a, Improved resolution of pipes with full waveform
inversion of common-offset GPR data using PEST: Geophysics, Submitted.
Jazayeri, S., and S. Kruse, 2016b, Full-waveform inversion of ground penetrating radar (GPR) data using
PEST (FWI-PEST method) applied to utility detection: 86th Annual International Meeting, SEG,
Expanded Abstracts, http://dx.doi.org/10.1190/segam2016-13878165.1.
© 2017 SEG Page 5144

Kalogeropoulos, A., J. van der Kruk, J. Hugenschmidt, S. Busch, and K. Merz, 2011, Chlorides and
moisture assessment in concrete by GPR full waveform inversion: Near Surface Geophysics, 9,
277–285, http://dx.doi.org/10.3997/1873-0604.2010064.
Klotzsche, A., J. van der Kruk, J. Bradford, and H. Vereecken, 2014, Detection and identification of high
contrast layers with limited lateral extent using an amplitude analysis approach and crosshole
GPR full-waveform inversion: Synthetic and experimental data: Water Resources Research, 50,
6966–6985, http://dx.doi.org/10.1002/2013WR015177.
Klotzsche, A., J. van der Kruk, N. Linde, J. Doetsch, and H. Vereecken, 2013, 3D characterization of
high-permeability zones in a gravel aquifer using 2D crosshole GPR full-waveform inversion and
waveguide detection: Geophysical Journal International, 195, 932–944,
http://dx.doi.org/10.1093/gji/ggt275.
Klotzsche, A., J. van der Kruk, G. A. Meles, J. A. Doetsch, H. Maurer, and N. Linde, 2010, Full-
waveform Inversion of crosshole ground-penetrating radar data to characterize a gravel aquifer
close to the Thur River, Switzerland: Near Surface Geophysics, 8, 631–646,
http://dx.doi.org/10.3997/1873-0604.2010054.
Lambot, S., E. C. Slob, I. van den Bosch, B. Stockbroeckx, B. Scheers, and M. Vanclooster, 2004,
Estimating soil electric properties from monostatic ground-penetrating radar signal inversion in
the frequency domain: Water Resources Research, 40, W04205,
http://dx.doi.org/10.1029/2003WR002095.
Tarantola, A., 1984, Inversion of seismic reflection data in the acoustic approximation: Geophysics, 49,
1259–1266, http://dx.doi.org/10.1190/1.1441754.
Virieux, J., and S. Operto, 2009, An overview of full-waveform inversion in exploration geophysics:
Geophysics, 74, no. 6, WCC127–WCC152, http://dx.doi.org/10.1190/1.3238367.
Wang, R., F. Xu, and X. Zhou, 2015, Blind deconvolution of seismic data based on the Spearman’s rho:
Journal of Computer and Communications, 3, 20–26, http://dx.doi.org/10.4236/jcc.2015.33004.
Young, R. A., Z. Deng, and J. Sun, 1995, Interactive processing of GPR data: The Leading Edge, 4, 275–
280, http://dx.doi.org/10.1190/1.1437130.
© 2017 SEG Page 5145

Stochastic multi-parameter coupling model in GPR application
Nan Huai1*, Zhaofa Zeng1,2, Jing Li1,2, Qi Lu1,2
1
College of geo-exploration Sci. &Tech, Jilin University, China.
2
Key laboratory of Applied Geophysics, Jilin University, China.
Summary inhomogeneous medium, and establish an accurate

mathematical relationship between electrical parameters
Geophysical method can be regarded as an indirect and dielectric characteristics (Zhu et al. 2003; Zhanov et
evaluation tool which can infer the intrinsic geological al. 2008). Accordingly, this method not only has high
properties of the target based on the recorded data which accuracy, but also can reflect the relationship between
is caused by the difference of physical parameters geologic and dielectric parameters.
(velocity, density, permittivity, conductivity and so on).
In fact, there is a certain coupling relationship between In this paper, based on the theory of effective medium
the physical attributes and the geological parameters and the coupling relationship of different parameters, we
(porosity et al). Therefore, making the geophysical establish the multi-parameter coupling stochastic media
parameters associate with the hydrogeologic properties models. Analysis shows this kind of model can associate
not only can improve the multi-parameter coupling abnormal parameters of geophysical detection with some
relations of geophysical model, but also can increase the hydrogeological parameters, and can fully reflect both
interpretation precision. In this paper, based on the microscopic changes and macroscopic properties of
theory of stochastic effective media, we use the Pride underground structure attributes.
model, and hybrid models of Hanai-Bruggeman and
Maxwel-Garnettto to deduce the coupling equations of Methodology
permittivity, conductivity and permeability parameters,
and build multi-parameter coupling stochastic media Conventional methods mainly use some models related
model. On the basis of the above study, we further to the empirical formula to describe the relationship
analyze the influence of various parameters on the between dielectric parameters and the moisture content,
distribution of random parameters model. such as the Topp formula and the CRIM model. In this
paper, we adopt the effective dielectric constant model
Introduction proposed by Peplinski et al. (1995), and compare the
relations between dielectric constant and the changes of
For the random inhomogeneous media model, a large water content with the above three models in Figure 1.
amount of effort was spent on improving it. Deutsch et With the increase of water content, the variation
al. (1996) introduced the details of definition and amplitude and trend of dielectric constant of three
modeling methods of stochastic media model. Varadan models are obviously different. The Topp formula and
et al. (1989) studied the process of scattering and CRIM model are both empirical fitting formulas under
attenuation of elastic waves in random media. Korn some idealized conditions. Therefore, there is a certain
(1993) studied the propagation of seismic waves in error in estimating the water content or dielectric
random media, and gave a statistical description of constant through these formulas (Giroux et al. 2010). By
stochastic media models. Ikelle et al. (2012) carried out comparison, the effective media model shown in black
2D forward simulation on random media with ellipsoidal solid line can more accurately describe the coupling
autocorrelation function. equations of different parameters.
Traditional modeling of stochastic media mainly

describes the random distribution of media properties
through various mathematical functions. However,
numerous problems still have not been solved. Although
mathematical functions can express an arbitrary random
distribution of the model, it is essentially a random
distribution of values, and the data itself lacks practical
physical meaning. What’s more, the parameters of this
kind of random media model are independently and
randomly distributed. Kirkpatrick proposed the effective
medium theory, which is an effective method to the
description of non-uniformity in a small scale (Lagarkov
et al. 1996; David et al. 1990). The basic purpose of the Figure 1: Curves of dielectric constant changing with the water
effective medium theory is to obtain the effective content in different empirical formula.
electric parameters from the distribution of
© 2017 SEG Page 5146

1. Coupling equation of porosity and conductivity The Cem and Cos in eq. 1 can be described as follows,
respectively:
Due to the existence of clay and other components, the L
ezl
Archie formula cannot be used to estimate the Cem  2d  (ezl )2 bl N L [exp( )  1], 
l 1 2kT
conductivity parameters directly. Many models have
been proposed taking the effects of surface conductivity where d is the length of Debye, it represents the
into consideration. We mainly refer to the model measured thickness of double-layer diffusion factor, its
proposed by Pride, where the conductivity σ can be definition is as follows:
described as follows (Pride, 1994; Tabbagh et al. 2002): 1 L
(ez )2 Nl
 l , 
 2[Cem  Re(Cos (w))]  d 2 l 1  0 f kT
  [ f S n  ],
  where  0 and  f are the permittivity of vacuum and fluid,
where  is the porosity field,  is the conductivity that
respectively, and k  1.38 1023 J / K is the Boltzmann
ignores the surface,  f is the conductivity of fluid, S is
coefficient. T is the Kelvin temperature,
saturation extent, n is the saturation exponent, Cem is   0.008  0.026log10 S , is electric potential.
the residual conductivity related to the electromigration ( 0 f  )2 P 2i 3/ 2 d 1
of double-layer particles, Cos is the frequency-dependent Cos  (1  ) , 
2d f P
conductivity induced by double-layer particles through
f
where P  8kTd 2  LNl [exp( ezl )  1] ,  
2
the migration of induced current. , w is
 0 f  l 1 2kT w f
The conductivity of fluid  f can be calculated by L ions: angular frequency, and  f is the density of fluid.
L
 f   (ezl )2 bl Nl , 
l 1
2. Coupling equation of porosity and dielectric constant
19
where zl is the atoms, e  1.6 10 C is the exchanged
Druchinin proposed a model to calculate the dielectric
values of electric field ， Nl  1000sNa is the constant of mixed soil, considering the four components
concentration of ions, where s is the salinity of soil medium: air, moisture, clay and soil. The volume
( mol / l ), bl  1/ (6 f Rl ) represents the fluidity, fractions of each component are: a , w , c , s . The
Heine-Borg formula only considers the three-phase
where Rl  0.116nm is effective radius of ions.  f is
medium, consisting of minerals, air, and water.
fluid viscosity related to the temperature (Li, 2014):
when T  0  20C : The Heine-Borg formula based on the effective medium
1301 theory is given as follows (Zeng, et al. 2015):
 f  100exp[2.303(
998.333  8.1855(T  20)  0.00585(T  20)2 i   m  h w
3.30233)],
1  i  ( )( ) , 
i   h  m
 Where  h is the dielectric constant of the base,  i is the
when T  20  100C :
dielectric constant of inclusion, i is the volume
1.3272(20  T )  0.001053(T  20) 2
 f  100exp[2.303( )],  fraction of three-phase medium, w  1/ 3 .
T  105
Before deducing the parameter  in eq. 1, the surface However, eq.11 cannot be solved directly, therefore, we
coefficient  is defined as follows (Pride, 1994): use minimum-norm of Powell method (Giroux, 2008):
i   m  h w

Sp
 |1  i  ( )( ) | . 
Vh
, i   h  m
where S p represents the surface pores, Vh is the volume, According to the Maxwell-Garnett model, the mixed
dielectric constant can be described as follows (Giroux,
Vp 2010):
according to the definition of porosity,   , eq. 5
i
V
3
 j  x, y, z ( i   h ) / ( h  N j ( i   h ))
can be rewritten as eq. 6: m  h  h ,
 
  Cl c  (1  Cl ) s , 1  i  j  x , y , z N j ( i   h ) / ( h  N j ( i   h ))
3
where  c and  s represent the surface of pure clay and

pure soil, respectively, then according to the definition where N j is the depolarization factor in j direction
of special surface and porosity,   2Vp / S p . On the basis
(Press, 1992). In addition, we also take into account of
of above mentioned derivation: the influence of frequency and adopt the Cole-Cole
2 p
 .  model.
 (1   )
© 2017 SEG Page 5147

0  i We also discuss the distribution features of 3D random
    ,  medium field, using the 3D model of porosity field as
1  (i ) q

shown in Figure 4. The detailed description of the model
when the frequency is zero,    0 , which is related to data is provided in the references (Bayer, 2011). The
temperature: distribution of porosity is in the range of 0.1-0.3, an
 0 (T )  87.74  0.40008T  9.398 104 T 2  1.41106 T 3 , abnormality of layered porosity can be observed in the
 Figure 4. Through the above model, we can derive the

field of conductivity and relative permittivity,
when the frequency is infinite,     ,  is the angular
respectively, as shown in Figure 5.
frequency,  is relaxation time, always constant, and i
is the imaginary unit. The conductivity values in Figure. 5(a) change in the
range of 0.01 S/m-0.025 S/m, and the values of relative
Examples permittivity are within the range of 5-11. In comparison
with the porosity field, we can easily find the
We use the 2D and 3D stochastic porosity fields to distribution of conductivity and relative permittivity are
derive the corresponding coupling permittivity, both of high similarity with it. In summary, the
conductivity, and the distribution of other hydrogeological parameters can be well coupled with the
hydrogeological parameters. geophysical parameters, which make the study on
random medium distribution more pragmatic.
Furthermore, it is of great significance to numerical
simulation and parameter inversion
(a) (b)
Figure 2: 2D random field model of primary porosity. (a. Model
I. b. Model II)
Figure 2 shows the 2D random distribution field of

porosity model. With the above-mentioned coupling
relationship equation, 2D random field of relative Figure 4: 3D random field model of primary porosity.
permittivity and conductivity model are shown in Figure
3. It can be seen that the derived relative permittivity and
conductivity are both in good agreement with the
original porosity. Furthermore, the region which changes
greatly in the porosity field is also well represented in
the distribution of relative permittivity and conductivity.
(a)
(a)
(b)
Figure 5: 3D random model of conductivity and relative
permittivity.
(b)
Figure 3: 2D stochastic effective media model of relative
permittivity and conductivity. (a. Model I. b. Model II)
© 2017 SEG Page 5148

(a) (b) (c)
(d)
Figure 6: 3D schematic diagrams of detection results of 3D complex soil medium model. (a) 3D stereogram. (b) 3D slices in different
directions. (c) 3D slices of measuring lines. (d) Plane distribution of 8 measuring lines.
microscopic changes of the target medium, and helpful

In order to further verify the feasibility of the numerical to establish the coupling relations between the
simulation of multi-parameter coupling media model, we parameters of the media and the geophysical response.
use the 3D stratum model shown in Figure 5(b) for
numerical calculation by 3D finite difference method Acknowledgments
based on CFS-RIPML absorbing boundary (Li et al.,
2010 and 2012). The central frequency of the antenna is This work is supported by the National Key Research
set to 300MHz, the grid size is 321*201*141, and the Development Program: 2016YC06001104 and the
minimum mesh size is 0.02 m. Furthermore, we arrange Nature Science Foundation of China ((41574097 and
10 evenly spaced lines along the x-direction, exploring 41504083).
under the common offset observation mode. Figure 6
can clearly show the overall distribution of the model.
Figure 6(b), 6(c) and 6(d) show the slice charts of
exploration results of different directions and different
depth, which contributes to the comparison of different
detail features of abnormal distribution in the model. For
example, the bending anomaly at the depth of 25m in the
model is well reflected in the corresponding depth of the
simulation results. In addition, the high value areas of
relative dielectric constant at the depth of 100 m can be
also reflected in Figure 6(c).
Conclusions
In this paper, Firstly, we derive the coupling equations

between the porosity and dielectric constant,
conductivity, respectively. Then, we build several 2D
and 3D multi-scales and multi-fields coupling random
porosity models, based on the random mathematical
function and multi-parameter coupling equation, and
further deduced the coupling stochastic media model of
dielectric constant and conductivity. We also carry out
the 3D GPR numerical calculation and analyze the
corresponding wave field characteristics of the multi-
parameter coupling random medium. Analysis indicates
that this kind of multi-parameter coupling model can
clearly describe the macroscopic properties and
© 2017 SEG Page 5149

EDITED REFERENCES
REFERENCES
Bayer, P., P. Huggenberger, P. Rendard, and A. Comunian, 2011, Three-dimensional high resolution
fluvio-glacical aquifer analog: Part 1: filed study: Journal of Hydrology, 405, 1–9,
http://dx.doi.org/10.1016/j.jhydrol.2011.03.038.
David, C., Y. Gueguen, and G. Pampoukis., 1990, Effective medium theory and network theory applied
to the transport properties of rock: Journal of Geophysical Research Solid Earth, 95, 6993–7005,
http://dx.doi.org/10.1029/JB095iB05p06993.
Deutsch, C. V., and L. Wang., 1996, Hierarchical object-based stochastic modeling of fluvial reservoirs:
Mathematical Geosciences, 28, 857–880, http://dx.doi.org/10.1007/BF02066005.
Giroux, B., and M. Chouteau, 2008, A hydrogeophysical synthetic model generator: Computer &
Geosciences, 34, 1080–1092, http://dx.doi.org/10.1016/j.cageo.2007.11.006.
Giroux, B., and M. Chouteau, 2010, Quantitative analysis of water-content estimation errors using ground
penetrating radar data and a low-loss approximation: Geophysics, 75, 241–249,
http://dx.doi.org/10.1190/1.3464329.
Ikelle, L. T., S. K. Yung, and F. Daube, 2012, 2-D random media with ellipsoidal autocorrelation
functions: Geophysics, 58, 1359–1372, http://dx.doi.org/10.1190/1.1443518.
Korn, M., 1993, Seismic wave in random media: Journal of Applied Geophysics, 29, 247–269,
http://dx.doi.org/10.1016/0926-9851(93)90007-L.
Lagarkov, A. N., and A. K. Sarychev., 1996, Electromagnetic properties of composites containing
elongated conducting inclusions: Physical Review B, 53, 6318–6336,
https://doi.org/10.1103/physrevb.53.6318.
Li, J., Z. F. Zeng, L. Huang, F. S. Wu, L. Huang, 2010, Three dimensional high-order FDTD simulation
for GPR: Chinese Journal of Geophysics, 53, 974–981.
Li, J., Z. F. Zeng, L. Huang, and F. S. Liu, 2012, GPR simulation based on complex frequency shifted
recursive integration PML boundary of 3D high order FDTD: Computers & Geosciences 49,
121–130, http://dx.doi.org/10.1016/j.cageo.2012.06.020.
Li, J., 2014, Ground penetrating radar detection and parameter inversion in stochastic effective media:
PhD Thesis, Jilin University.
Peplinski, N. R., F. T. Ulaby, and M. C. Dobson, 1995, Dielectric properties of soils in the 0.3-1.3-GHz
range: IEEE Transactions on Geoscience & Remote Sensing, 33, 803–807,
http://dx.doi.org/10.1109/36.387598.
Press, W., B. Flannery, S. Teukolsky, and W. Vetterling, 1992, Numerical Recipes in C: the Art of
Scientific Computing: Cambridge University Press, 104, 95–96.
Pride, S., 1994, Governing equations for the coupled electromagnetics and acoustic of porous media:
Physical Review B: Condensed Matter, 50, 15678–15696,
https://doi.org/10.1103/physrevb.50.15678.
Tabbagh, A., C. Panissod, R. Guerin, and P. Cosenza, 2002, Numerical modeling of the role of water and
clay in soils’ and rocks’ bulk electrical conductivity: Journal of Geophysical Research
Atmospheres, 107, 2318–2326, http://dx.doi.org/10.1029/2000JB000025.
Varadan, V. K., Y. Ma, and V. V. Varadan, 1989, Scattering and attenuation of elastic waves in random
media: Pure and Applied Geophysics, 131, 577–603, http://dx.doi.org/10.1007/BF00876265.
© 2017 SEG Page 5150

Zeng, Z. F., X. Chen, J. Li, L. N. Chen, Q. Lu, and F. S. Liu, 2015, Recursive impedance inversion of
ground-penetrating radar data in stochastic media: Applied Geophysics, 12, 615–625,
http://dx.doi.org/10.1007/s11770-015-0514-0.
Zhanov, M. S., 2008, Generalized effective medium theory of induced polarization: Geophysics, 73,
F197–F211, http://dx.doi.org/10.1190/1.2973462.
Zhu, P. Y., A. K. Fung, and K. W. Wong, 1987, Effective propagation constants in dense random media
under effective medium approximation: Radio Science, 22, 234–250,
http://dx.doi.org/10.1029/RS022i002p00234.
© 2017 SEG Page 5151


Practicing S-wave reflection seismology with “P-wave” sources: concepts, principles, and overview
Bob A. Hardage*, Bureau of Economic Geology, Jackson School of Geosciences
Summary
This paper introduces a special session at the 2017 SEG Annual Meeting titled: Practicing S-wave reflection seismology with P-
wave sources. In this session, speakers will focus on a low-cost, effective, land-based technology that allows all possible P-wave
modes, all possible S-wave modes, and all possible converted modes to be utilized across surface conditions where standard S-
wave sources (e.g., horizontal vibrators) cannot be used to generate illuminating S wavefields (e.g., swamps, marshes, dense timber,
jungles, and rough mountainous terrains). The unique aspect of this new approach to full-elastic imaging is that the data that are
used are generated by common, widespread, low-cost sources that are traditionally referred to as “P sources”. Some definitions and
basic principles need to be explained at the onset of the session to assist attendees in understanding the data examples that will be
discussed by session speakers. That overview is the objective of this introductory paper.
Introduction
All onshore seismic sources can be segregated into two generic classes. Class-1 sources are vertical-displacement sources. Vertical-
displacement sources include vertical vibrators, vertical impacts, and buried explosives. Any source in this group is traditionally
called a P-wave source. These “P sources” are the sources that will be utilized in the full-elastic imaging discussed in this technical
session. Class-2 sources are horizontal-displacement sources and include horizontal vibrators and horizontal impacts. These
sources are traditionally called S-wave sources. Although Class-2 sources (S sources) generate excellent illuminating S-wavefields
and need to be available to the geophysical industry, the purpose of this technical session is to demonstrate that Class-1 sources (P
sources) are lower-cost alternatives to Class-2 sources for purposes of S-wave refection seismology.
Two types of illuminating wavefields, P illumination and SV illumination, are produced by each source in these two generic-source
classes. In some cases, one of these two illuminating wavefields may be too weak to be effective for reflection imaging of deep
targets. An example would be the P illumination produced by horizontal vibrators. If an illuminating P wavefield, or an illuminating
SV wavefield, is produced immediately at a source station, that wavefield will be referred to as a direct-P, or as a direct-SV,
wavefield. If a downgoing illuminating P or SV wavefield is created by mode conversion at an interface remote from a source
station, that wavefield will be called a converted-P or a converted-SV wavefield.
Theoretical concepts
Two concepts form the physics of the technology that will be discussed in this technical session. The first concept is that all Class-
1 vertical-displacement sources (i.e. traditional P sources) produce both a robust direct-P and a robust direct-SV illuminating
wavefield (Figure 1a). This fundamental wave physics has been ignored for decades for inexplicable reasons, and reflection
wavefields produced by only the direct-P mode generated by these sources have been used in reflection seismology. This session
will stress that important seismic images and valuable seismic attributes have been overlooked because geophysicists have not
attempted to utilize the downgoing direct-SV illuminating wavefields produced by Class-1 sources. The second concept is that the
direct-P and direct-SV wavefields produced by “P sources” cause vertical geophones to always record two independent and
interlaced reflection wavefields (a P-P wavefield and a SV-P wavefield), and similarly, horizontal geophones to also record two
independent and interlaced wavefields (a SV-SV wavefield and a P-SV wavefield), as illustrated in Figure 1b.
Expanded imaging options provided by P sources were first discussed in geophysical literature in 2014 (Hardage and Wagner,
2014a, 2014b; Hardage, et al., 2014). Before these papers were published, there was not a single example in the literature of a SV-
P image or a SV-SV image produced by a P source and recorded by surface receivers. The theme of this session will be to illustrate
new information that becomes available for geologic interpretation when P-source data are considered from the points of view that:
(1) two downgoing wavefields (a direct-P wavefield and a direct-SV wavefield) are produced simultaneously by P sources, and (2)
two overlapping reflections (one produced by the direct-P wavefield and one produced by the direct-SV wavefield) are present in
P-source vertical-geophone data and also in P-source horizontal-geophone data.
P and S radiation patterns produced by P sources
The basic physics that vertical-displacement sources produce both direct-P and direct-SV illuminating wavefields has been in the
literature since the 1950s (Miller and Pursey, 1954). Rather than just accepting the theory that a P source produces two,
simultaneous, illuminating wavefields, researchers at the Bureau of Economic Geology (BEG) performed field tests that measured
the direct-P and direct-SV illuminating energy produced by all commercially available Class-1 and Class-2 seismic sources listed
above. Examples of test results for only Class-1 P sources are exhibited in Figure 2. These results show, that at BEG’s Devine Test
Site, all P sources generate much more illuminating SV energy than illuminating P energy. Because there is moderate azimuthal
anisotropy at the Devine Test Site, robust SV energy appears even on transverse geophones (blue data points, Figure 2). These raw,
digital, SEGY-format, BEG test data acquired for all Class-1 and Class-2 sources have been shared with research groups at 7
1

© 2017 SEG Page 5152

S reflection seismology with P sources
universities for independent analysis, and are available to other universities and investigators on request. Additional field-test
measurements of P and S radiation patterns produced by all types of sources need to be done by other researchers across different
surface conditions, and in various subsurface propagation media, to determine if P sources exhibit direct-P and direct-SV
illumination physics in those test conditions that are similar to, or different from, what we observed at our test site (Figure 2).
Full-elastic imaging with P sources
When BEG initiated the concept of practicing S-wave reflection seismology with P-sources in 2010, our objective was to develop
a lower cost option for acquiring full-elastic seismic reflection data (i.e., acquiring all possible P-mode reflections and all possible
S-mode reflections). An example of full-elastic imaging with P sources that was developed in our first investigation is illustrated
in Figure 3. This 4-panel figure exhibits two common-midpoint images (P-P and SV-SV) and two converted-mode images (SV-P
and P-SV). The source that generated these legacy data (acquired approximately 8 years earlier) was an array of 3 inline vertical
vibrators. No effort was made to segregate images into fast and slow modes. Also the images in Figure 3 represent products made
early in the data-processing strategy when data were being studied in super-bin form. We claim these images are the first-ever
examples of full-elastic imaging with P sources. Please inform the author if we at BEG have missed important evidence contrary
to this claim.
Switch from full-elastic imaging to imaging with vertical-geophone data only
As soon as BEG began to show the examples in Figure 3 to peers and colleagues, we received strong encouragement to first
concentrate on developing technology that produced SV-P reflections from vertical-geophone data. The motivation for this repeated
recommendation was that once SV-P imaging technology is developed that utilizes only P sources and vertical geophones, then
hundreds of 1000s of square kilometers of legacy P-wave data can be reprocessed to acquire an S-mode image and S-wave attributes
without having to expend funds for new data acquisition done specifically for the purpose of acquiring S-wave information. We
listened to this advice, and as a result, our objective of practicing S-wave reflection seismology with P sources has, to date, been
focused on SV-P imaging. For this reason, the principal information presented in this technical session will be on utilizing the SV-
P mode embedded in P-source, vertical-geophone data, not on performing full-elastic imaging with P sources and 3C geophones
(Figure 3). SV-SV imaging with P sources and 3C geophones will rise to the forefront of our interest soon and will probably begin
to dominate our research effort.
SV-P stacking velocities from CMP-based, constant-NMO velocity stacks of vertical-geophone data
We have found that valuable information about SV-P reflections comes from data products that are built during conventional P-P
data processing, particularly when data processing is focused on determining P-P stacking velocities. Wiggle-trace displays of
constant-velocity (CV), normal-moveout (NMO) stacks of vertical-geophone data such as are commonly created to determine P-P
stacking velocities will be used to illustrate a valuable analysis procedure practiced at BEG which allows a person to get an early,
first-look at SV-P reflections. The concept is illustrated in Figure 4, which shows a simplified version of a panel of CV, NMO-
corrected, CDP trace gathers similar to what many data processors make in order to determine P-P stacking velocities.
In this display, an obvious P-P reflection BPP is assumed to be in velocity panel VPP at time TPP. Work done by Tessmer and Behle
(1988) and Iverson et al. (1989) show that the stacking velocity VPS for the P-SV mode that is depth-equivalent to P-P reflection
BPP is:
(1) (VPS)2 = (VPP)2 / A,
where A is the average VP/VS velocity ratio for the propagation medium. It is correct to say that VSP is approximately equal to VPS
unless velocity layering is particularly complex. Thus equation 1 is written in Figure 4 using VSP, not VPS, as the equation was
originally developed. The end result is an estimate of the velocity panel VSP where the SV-P reflection BSP that is depth-equivalent
to P-P reflection BPP should be located. The simple equation,
(2) TSP = 0.5 (1 + A) TPP,
then indicates the time coordinate in this VSP velocity panel where SV-P reflection BSP should appear. A reasonable-size rectangle
abcd in velocity-time data space is centered on this calculated time coordinate to serve as a search area in which to identify the
most likely candidate for a SV-P reflection event.
This simple velocity analysis, when applied to legacy P-P data, provides two valuable pieces of information needed for extracting
SV-P reflections from vertical-geophone data. First, the analysis confirms whether or not SV-P reflections are present in the data.
This insight is invaluable so that time is not wasted attempting to retrieve SV-P reflections from legacy data where short trace
lengths, inappropriate vibrator sweeps, low fold, and other factors involved in earlier-day, P-P data-acquisitions have not allowed
2

© 2017 SEG Page 5153

S reflection seismology with P sources
SV-P reflections to survive. Second, a SV-P stacking velocity is created that is correct to first-order, which allows SV-P data
processing to proceed rapidly. Third, if SV-P CMP reflections discovered by this analysis are reasonable quality, confidence is
built that SV-P reflection quality will only improve when the same reflections are processed according to ACP binning and imaging
principles, not according to the CMP binning and imaging principles used in the P-P stacking velocity analysis.
A real-data example is provided as Figure 5 in which SV-P reflections are identified in CMP trace gathers of vertical-geophone
data after constant-velocity NMO corrections are applied. Only two P-P reflections A and B will be utilized. Using the equations
from Figure 4, the positions of SV-P reflections that are depth-equivalent to A and B are calculated and shown as green circles.
These seismic data were acquired with 3C geophones, and the green curve is the stacking-velocity function for the P-SV data
extracted from the companion horizontal geophones at this same CDP. Note that the positions of the calculated SV-P reflections
coincide with the P-SV velocity curve. As a point of curiosity, the SV-SV stacking velocity function determined at the same CDP
is added to the CV panel, and predicted positions of SV-SV reflections that are depth-equivalent to P-P reflections A and B are
shown as yellow circles. It is rare to find meaningful SV-SV reflection in vertical-geophone data. However, BEG investigators
have found interesting possibilities in some instances, such as the yellow circle that correlates with P-P reflection B in this case.
Conclusions
Irrefutable evidence has been amassed showing that P sources illuminate geology with both direct-P and direct-SV wavefields.
This basic source physics allows full-elastic wavefield imaging to be practiced with P sources if data are recorded by 3C geophones.
The physics also allows an SV-mode image and SV attributes to be created through the SV-P mode if P-source data are recorded
with only vertical geophones. In theory, an SV-P image is equivalent to the P-SV image that is acquired using a P source and 3C
geophones. The image equivalence of P-SV and SV-P data has been established by Li and Hardage (2015). It is a mixture of sadness
about lost opportunities, and excitement about future potential, when one realizes that untapped S-wave information lays dormant
in hundreds of 1000s of square kilometers of legacy P-P data that are preserved in seismic data libraries around the globe. Because
of the immense value of revisiting legacy P-P data and generating SV images and attributes, the initial focus of developing S-wave
reflection seismology with P sources by focusing on only SV-P imaging is well justified. However, we who speak at this session,
and others who will join in developing this new technology, plan to expand P-source applications into full-elastic imaging. Our
objective is not to abandon traditional S-wave sources (horizontal vibrators and inclined impacts), but to show that these Class-2
sources are not required for illuminating geology with direct-S modes unless a person prefers to use them.
Figure 1. (a) Key concept #1 – common P-wave sources produce both a direct-P illuminating wavefield and a direct-SV
illuminating wavefield. (b) Key concept #2 – as a result of this dual, direct-wavefield illumination, vertical geophones record two
independent reflection wavefields, and so do horizontal geophones.
Figure 2. Direct-P and direct-SV radiation patterns produced by common Class-1 P sources at the Devine Test Site: (a) vertical
vibrator, (b) vertical impact, and (c) buried explosive.
3

© 2017 SEG Page 5154

Figure 3. First-ever example of full-elastic wave imaging with a P source. Images are (a) P-P, (b) P-SV, (c) SV-P, and (d) SV-
SV. The source was an inline array of three vertical vibrators. The receivers were a single 3C geophone at each receiver station.
Figure 4. Simplified view of a constant-velocity panel constructed from common-depth-point (CDP) traces for purposes of
determining P-P stacking velocities. The SV-P reflection BSP that is depth equivalent to P-P reflection BPP should be located
inside search area abcd centered on velocity VSP (defined by equation 1) and image time TSP (defined by equation 2).
Figure 5. Real-data example of practicing the principles illustrated in Figure 4. Source was an array of vertical vibrators.
Receivers were vertical geophones of 3C geophone stations. The calculated positions of SV-P stacking velocities (green circles)
fall precisely on the companion P-SV stacking velocity function (green curve) determined from horizontal-geophone data.
4

© 2017 SEG Page 5155
EDITED REFERENCES
REFERENCES
REFERENCES
Hardage, B. A., and D. Wagner, 2014a, Generating direct-S modes with simple, low-cost, widely
available seismic sources: Interpretation, 2, SE1–SE15, http://doi.org/10.1190/INT-2013-0095.1.
Hardage. B. A, and D. Wagner, 2014b, S-S imaging with vertical-force sources: Interpretation, 2, SE29–
SE38, http://doi.org/10.1190/INT-2013-0097.1.
Hardage, B. A., D. Sava, and D. Wagner, 2014, SV-P: An ignored seismic mode that has great value for
interpreters: Interpretation, 2, SE17–SE27, http://doi.org/10.1190/INT-2013-0096.1.
Iverson, W. P., B. A. Fahmy, and S. B. Smithson, 1989, VPVS from mode converted P-SV reflections:
Geophysics, 54, 843–852, http://doi.org/10.1190/1.1442713.
Li, Y., and B. A. Hardage, 2015, SV-P extraction and imaging for far-offset vertical seismic profile data:
Interpretation, 3, SW27–SW35, http://doi.org/10.1190/INT-2015-0002.1.
Miller, G., and H. Pursey, 1954, The field and radiation impedance of mechanical vibrators on the free
surface of a semi-infinite isotropic solid: Proceedings of the Royal Society of London, Series A,
223, 521–541, http://doi.org/10.1098/rspa.1954.0134.
Tessemer, G., and A. Behle, 1988, Common reflection point data-stacking technique for converted waves:
Geophysical Prospecting, 36, 671–688, http://doi.org/10.1111/j.1365-2478.1988.tb02186.x.
© 2017 SEG Page 5156

AVO and interpretive processing of Sv-P reflections co-existing with P-P reflections on
P-wave legacy data
Mike Graul*, Texseis, Inc.
Summary lower with Vp/Vs ratios ranging to as high as 10 in the very

near surface. The result is that the source statics for the Sv-
The demonstrated presence of Sv-P reflections on regular P reflections are very large and problematic. The shear
P-wave data, with P-wave sources (vertical vibrators, uphole time for buried explosives is not available.
explosives, vertical impact) and vertical receivers, opens Refraction statics are difficult to implement owing to the
the door for converted wave processing to confirm and dominance of P-wave refractors and other surface waves
augment normal P-P processing. moving at velocities like that of the shear refractor.
Introduction Compounding the problem, and making the use of reflection

or residual statics very difficult, is the fact that the common
From Miller and Pursey (1954), who showed that the mid-point gather (CMP) which identifies a collection of
radiation pattern from vertical vibrating sources at the traces with source and receiver pairs (S-R) having a
surface contained a rich amount of direct Sv waves, through common mid-point on PP reflection data, will not serve for
a wide assortment of papers and lectures by Bob Hardage Sv-P data even for flat dip, as having S-R commonality. The
and his colleagues, including numerous field experiments, problem is illustrated in Figure 1, below.
there has been ample proof of the presence of direct Sv
waves from P sources and Sv-P reflections on P-wave data. . X
Despite the potential applications for exploration, very little
has been done to tap this rich and inexpensive source of S CMP R
valuable information. Most of the effort in the use of
converted waves has been done with P-Sv, generally
requiring multi-component receivers, or S-wave sources CCP
and multi-component receivers. In either of the latter cases, Sv-P
the cost factor mounts rapidly.
Z
The potential for comparative analysis of AVO or AVA of the
P-P
Sv-P and P-P reflections is a prominent application of the s p
data, and is examined here along with other subtler uses of
the dormant data. Vs   Vp
Processing Issues for Sv-P data

ACP
The essential difficulty with extraction of either P-P or Sv-P
data from legacy P-wave recordings is that these body
𝑠𝑖𝑛∅𝑠 𝑠𝑖𝑛∅𝑝
= = 𝑝, 𝑡ℎ𝑒 𝑟𝑎𝑦 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟
waves are superimposed in time and space along with the 𝑉𝑠 𝑉𝑝
ubiquitous overlap of surface waves, multiples, and other X = Source – Receiver Offset
classes of noise. s = incident angle Sv source
It is the processor’s responsibility to separate the modes p = reflection angle Sv-P
and present the Sv-P and P-P reflectivities in an  = incident and reflection angle P-P
interpretable format. CCP: Common Conversion Point: varies
with depth, Z, and offset, X.
The converted wave (Sv-P) requires some extra effort in the CMP: Common Mid-Point
execution of standard processing steps. The goal is to ACP: Asymptotic Conversion Point
produce Pre-Stack Depth Migrated (PSDM) data for both
Sv-P and P-P which may be used independently for an Figure 1: The geometry relating Sv-P and P-P
enhanced interpretation. We will assume that the standard reflections, their respective reflection points and
processing for P-wave reflectivity has already taken place. angles in the subsurface.
Since the use of P-wave sources and receivers were in the
original acquisition and processing design, this would be a
reasonable assumption. It will be used as a reference It is important to note that in a CMP gather, commonly (and
often incorrectly) used for both residual reflection statics
The first issue is the Sv static corrections on the source and VNMO analysis, whole traces of varying offset are
side. While the very near surface weathering velocity, Vw, collected together. This is not the case for the Sv-P
and layer below it, Vd for datum or “replacement” velocity, converted wave. The simplifying assumption for P-P
are typically close to 2000 ft/s and 6000 ft/s respectively, reflections is that CMP  CDP, the common depth point or
the corresponding shear velocities may be significantly
© 2017 SEG Page 5157

AVO and interpretive processing of Sv-P reflections
analysis to converge acceptably to VSINT(z). The test of
common reflection point (CRP). This assumption is valid for acceptable is the registration of the P-P and Sv-P events
P-P reflections only when dealing with flat layering and one on their respective migrations.
dimensional velocity functions. For converted waves, the
reflection point is the conversion point and it will shift

laterally at each reflector depth (or travel time). The
configuration of this lateral shift is shown as the ACP curve
which asymptotically approaches a constant value as depth
increases. There is no way to obtain a valid equivalent of Sv-P
the CMP for converted waves. For residual statics, we are
forced to use only deep reflectors and short offsets.  - 1.8
The processor faces a similar problem in determining Sv-P
velocity. Here, however, the goal is a model of both Vs and
Vp throughout the volume of data. The initial stacking or
NMO velocity, derived from the unmigrated data, can be
estimated roughly using the techniques described by
Tessmer and Behle (1988) and Iverson, et al (1989). Their
method assumes an approximate Vp/Vs = , and yields an
estimate of Vsp from CMP velocity analyses using either
P
semblance plots or constant velocity stack panels.
While the estimated converted wave velocity, Vsp, in its rms

mode may provide a rough and un-migrated preview of the P-P
Sv-P reflection data, it serves no other useful processing
purpose at this point in the sequence.
Proper imaging of Sv-P data can be accomplished most

effectively using PSDM, requiring knowledge of Vs and Vp
in their interval velocity manifestations, VSINT(z) and
VPINT(z). With further information and calibration, we may
extend this concept to include anisotropy in APSDM.
The preliminary velocity analysis on CMP data may be used

as the starting point for time migration, PSTM. Using
iterative tomographic velocity updating, convergence is
achieved on a final VRMS usable for initial estimates of the
Sv-P
VPINT(z) to be used for P-wave PSDM. Again, the  - 2.0
application of iterative tomography brings about
convergence to an acceptable VPINT(z) at each P-wave
common image point (CIP), which may be regarded as
Z
located at each inline (IL), crossline (XL) intersection on a Figure 2:Synthetic example of registration of Sv-P and P-P
3D survey. We are now ready for determination of VSINT(z), reflections updating  in PSDM. [A] Sv-P PSDM event with
at the same locations, but in this case the CIP will play the initial  = 1.8; [B] P-P PSDM; [C] SV-P PSDM with corrected
role of a CCP, with different S-R offsets, as well.  = 2.0. All panels traces are shown at corresponding P-
wave angles at the same Common Image Point (CCP or
For interpretive purposes, the registration of the P-P and CDP), at or around true depth = Z.
Sv-P reflections is critical. Defining Vs and Vp as the
respective average velocities to depth Z, we may write,
(1) Z/Vs = Ts {Sv-wave vertical travel time} The reflections should appear flat and at the same depth
(2) Z/Vp = Tp {P-wave vertical travel time} (assumed to be correct on the P-P data). Initially, it is not
(3) Vp/Vs =  {the ratio at depth Z}. uncommon to set  to a constant or to use Castagna’s
Combining these we find that, Mudrock line, allowing depth variation.
(4) Tsp = Tp(1+) {Sv-P travel time}
(5) Vsp = 2Z/Tp(1+) {average Sv-P velocity}. Using the P-P data as the reference, the corresponding Sv-
P reflections must be identified and their depth-varying
The key information is the Vp/Vs ratio which ideally, but shifts, Z, measured. These steps are best incorporated in
rarely, comes from log measurements of sufficient number a depth varying correlation scheme, accounting for the
to define the 3D volume. It will still require iterative velocity anticipated polarity reversals as they would appear on the
stacks. These values coupled with the residual NMO on the
© 2017 SEG Page 5158

Sv-P PSDM gathers may then be used to estimate the comparing, on real data, the PSDM and the corresponding
required updates to (IL, XL, Z), using a least squared error section with Sv-P suppressed.
(LSE) iteration scheme. The corresponding average
velocities, VP (IL, XL, Z) and VS (IL, XL, Z), in the volume,
In turn, may be used to compute smoothed Interval P-P Sv-P

INPUT
velocities, VSINT (IL, XL, Z) and VPint (IL, XL, Z), for Sv-P PSDM Suppressed
migration. A B C
After P-P PSDM, It is common to find interfering curved
(residual moveout) events masking the p-wave reflections.
While often interpreted as multiples, they are demonstrably
more likely to be Sv-P converted waves. Similarly, on Sv-P
migrations, under-migrated P-P reflections appear,
exhibiting convex curvature and interference with the now
flattened Sv-P events. In both cases, it is beneficial to
suppress the interfering mode to facilitate pre-stack Sv-P P-P
analysis, including AVO operations on both modes. Figure INPUT PSDM Suppressed
3 illustrates the abundance of Sv-P reflections which
appear on legacy P-P data. D E F
P-P PSDM Gather
Sv-P
PSDM Suppressed
G H
Multiples? Unlikely.
Analysis of the
depth relationship,
amplitude, and
residual NMO
indicates these are
probably Sv-P
reflections Figure 4: The upper 3 panels [A – C] show the input, P-
P PSDM, and Sv-P suppression. Panels [D – F] are
similar but with Sv-P migration and P-P suppression.
Figure 3: Interference between migrated P-P events and Panel G is a real PSDM with H the Sv-P suppression.
converted Sv-P reflections are shown on a CIP gather of
the P-P PSDM migration.
AVO applications of Sv-P reflections

The pre-stack suppression of either mode draws on careful
application of iterative LSE subtraction of estimates of the The reflectivity of Sv-P waves, Rsp(P), differ significantly
offending mode. Modelled and real data examples are from those of the P-P reflectivity, Rpp(P), where P is the
shown in Figure 4. reflected P angle of both modes. Each depends on the
same rock property contrasts at the reflective boundary
The normal suppression provided by stacking is not enough (Vs, Vp, and ), but because of orthogonal particle
to attenuate the relatively high amplitude Sv-P waves often motions of P and S at the boundary, they manifest in
seen on the PSDM stacks. A special post stack technique different ways. By comparing the AVO [AVA] of the two
is combined with the pre-stack conditioning to provide an types, independent confirmation of the interpreted inversion
improved stack. An illustration of this is given in Figure 5, may be derived. This represents a major advancement in
© 2017 SEG Page 5159

the application of AVO to exploration and to reservoir evolving technology has barely been tapped. There is
characterizations. building evidence that other interpretational applications of
Sv-P waves may not only augment PP interpretation, but
The data is available on virtually all previously recorded P-
wave data. This went unrecognized for decades. The Vp =1627 ft/s Rpp
breakthrough came with the work of Bob Hardage who Vs = 2570 ft/s
demonstrated the presence of direct Sv waves from so  = -0.27 g/cc
called P-wave sources. The data is there for the taking, R
requiring only processing without new acquisition, Rpp
0 A P 
P-P PSDM Rsvp
Rsvp
- .2
0
P 30
Vp = -569 ft/s
Vs = 42 ft/s
R  = -0.27 g/cc Rpp
Rsvp
0
P 
B
Rpp
- .1 Rsvp
Sv-P Suppressed P
0 30
R Vp = -569 ft/s Rpp

Vs = 1042 ft/s
 = -0.27 g/cc
Rsvp C
Rsvp
Rpp
- .2 P
0 30 P 
Figure 6: AVO responses of Rsp and Rpp

Figure 5: In the upper P-P migration, examples of the approximations. [A] Type 2 reservoir; [B] top
interfering Sv-P reflections are indicated by the yellow brine sand; [C] Gas Sand.
arrows. The lower panel shows their suppression. may even surpass it. One such example is the Sv-P ability
to delineate fault interfaces.
Close approximations of the Sv-P and P-P amplitude
behavior as a function of incident angle are given by Aki and Conclusions
Richards (2009). Their equations have been used to
compute the Rsp and Rpp AVO responses shown in Figure While the processing of Sv-P data is fraught with difficulties,
6. Panels 6B and 6C compare a simulations brine of sand the rewards are many. A major positive factor is that there
and gas sand responses, respectively. The “bright spot” is no acquisition cost or specialized equipment required.
behavior of the PP reflection is overridden by the positive Ordinary P-wave sources and receivers are adequate for
gradient of the Rsp response denying any gas-related the recording and processing of both P-P and Sv-P
reflectivity. The negative gradient of the Rsp curve in case recordings. The removal of one mode from the other is
6C confirms the gas sand. mutually beneficial, and the interpretive use of both has
proven surprisingly beneficial. It is only surprising that these
Case 6A simulates a Type 2 reservoir response with direct Sv waves haven’t been used before.
confirming evidence by the Rsp AVO. The potential of this
© 2017 SEG Page 5160

EDITED REFERENCES
REFERENCES
Aki, K., and P. Richards, 2009, Quantitative seismology (2nd ed.): University Science Books, ISBN 978-
1891389-63-4.
Hardage, B. A., 2017, Practicing S-wave reflection seismology with “P-wave” sources: Concepts,
principles, and overview: 87th Annual International Meeting, SEG, Expanded Abstracts.
interpreters: Interpretation, 2, SE17–SE27, https://doi.org/10.1190/INT-2013-0096.1.
Hardage, B. A., and D. Wagner, 2014a, Generating direct-S modes with simple, low-cost, widely
available seismic sources: Interpretation, 2, SE1–SE15, https://doi.org/10.1190/INT-2013-
0095.1.
Hardage, B. A., and D. Wagner, 2014b, S-S imaging with vertical-force sources: Interpretation, 2, SE29–
SE38, https://doi.org/10.1190/INT-2013-0097.1.
Iverson, W. P., B. A. Fahmy, and S. B. Smithson, 1989, VPVS from mode converted P-SV reflections:
Geophysics, 54, 843–852, https://doi.org/10.1190/1.1442713.
Interpretation, 3, SW27–SW35, https://doi.org/10.1190/INT-2015-0002.1.
Miller, G., and H. Pursey, 1954, The field and radiation impedance of mechanical vibrators on the free
surface of a semi-infinite isotropic solid: Proceedings of the Royal Society of London, Series A,
223, 521–541, https://doi.org/10.1098/rspa.1954.0134.
Tessmer, G., and A. Behle, 1988, Common reflection point data-stacking technique for converted waves:
Geophysical Prospecting, 36, 671–688, https://doi.org/10.1111/j.1365-2478.1988.tb02186.x.
© 2017 SEG Page 5161

There is no pure P- or S-wave land seismic source
André J.-M. Pugin*, Geological Survey of Canada, and Oz Yilmaz, Anatolian Geophysical
rise to a significant P-wave radiation (the green circle), but

Summary: also can yield S-wave radiation in two parts --- the
We conducted a field experiment at a soil site near Ottawa supercritical component (the red lobes) associated with
and recorded 9-C seismic data using a hand-held hammer surface wave and the subcritical component (the blue lobes)
and a receiver cable with 48 3-C 28-Hz geophones at 0.75- associated with downgoing S-wave. Such radiation
m intervals. The receiver spread length is 35.75 m and the patterns have also been modeled by Hardage and Wagner
near-offset is 0.75 m. We recorded three triplets of shot (2014). These radiation patterns, contrary to common
records with the impact source in vertical, inline horizontal, understanding, demonstrate that a vertical-impact seismic
and crossline horizontal orientations. We identified several source can generate both P and S waves, and that
wave modes in the nine field records --- PP, PS, SP, and SS partitioning of the source energy depends on the Vp/Vs
reflections, in addition to refracted waves. We then ratio. The fact that a vertical-impact source can generate
significant S-wave energy has led to the idea of extracting
performed vertical sum of the three records associated with
SP reflections from conventional vertical-geophone data
each of the three different source orientations (vertical,
inline horizontal, and crossline horizontal) but with (Hardage, 2015).
common geophone orientation, and computed the
semblance spectra of the composite records. We
ascertained the wave modes based on the semblance peaks.
This field test led to several important observations
regarding characteristics of wave propagation in the near-
surface. First and foremost, based on our experience in
using impulsive and vibroseis sources, there is no pure P- or
S-wave land seismic source --- any source type can
generate any combination of wave modes. Second, a wave
mode may not be present in a record acquired with a given
source-receiver orientation which theoretically should give
rise to that mode, but can appear unexpectedly in a record
acquired with a given source-receiver orientation which
theoretically should not give rise to that mode. Third, the
combination of wave modes captured by a specific source-
receiver orientation depends on the Vp/Vs ratio. Finally,
these observations led to a realization that for a complete
representation of the wavefield propagating within the near-
surface, we need to record multicomponent data.
Introduction:
Consider a vertical-impact source applied to the free
surface associated with an elastic half-space. The equations
for P- and S-radiation patterns have been derived by Miller
and Pursey (1953). Using these equations, F. Hilterman
wrote a program to calculate the radiation patterns
(personal communication). Figure 1 shows the radiated
wave modes for two different Vp/Vs ratios --- typical of a
soil column (Figure 1a) and a rock column (Figure 1b).
Note that the vertical-impact source, which is mistakenly
defined as an ideal P-wave source, contrary to common Figure 1. Radiation patterns from a vertical-impact source
understanding, gives rise to a very small amount of P-wave for two cases of Vp/Vs ratio: (a) 4, and (b) 1.75.
radiation (the tiny green circle in Figure 1a); however, Field Experiment:
much of the source energy is consumed by the S-wave
radiation in the supercritical region (the red lobes) Inspired by the fact that a vertical-impact seismic source
contributing to surface waves. In the case of a lower Vp/Vs can generate both P- and S-waves, we conducted a field
ratio (Figure 1b), note that the vertical-impact source gives experiment at a soil site near Ottawa and recorded 9-C
© 2017 SEG Page 5162

There is no pure P- or S-wave source
seismic data using a hand-held hammer and a receiver cable bedrock as these patterns are less obvious on the H1- and
with 48 3-C 28-Hz geophones at 0.75-m intervals. The H2-geophone gathers.
receiver spread length is 35.75 m and the near-offset is 0.75

m. We recorded three triplets of shot records with the As the theory would predict, the PS-SP bedrock reflection
impact source in vertical (V), inline horizontal (H1), and at 150 ms occurs with its highest amplitude on the H1-
crossline horizontal (H2) orientations (Figure 2). Each geophone gather with a velocity of 550 m/s. It is a total
triplet consists of shot records acquired by common source surprise to observe that the highest amplitude of the PP
orientation and three geophone orientations (V: vertical, bedrock reflection is not on the V-geophone but on the H1-
H1: inline horizontal, and H2: crossline horizontal). In geophone gather at approximately 40 ms, this reflection is
Figure 2, each column represents a triplet acquired by clearly visible on H1,H1 (Figure 2). As expected, the H2-
common source, and each row represents a triplet acquired geophone gather does not display coherent amplitude peaks
by common geophone orientation. for P-wave.
Data Analysis:
Conclusions:
We performed vertical sum of the three records associated
with each of the three different source orientations but with This field test led to several important observations
common geophone orientation, and computed the NMO regarding characteristics of wave propagation in the near-
semblance spectra of the composite records with common surface.
geophone orientation (Figure 3). Based on velocities, (1) There is no pure P- or S-wave land seismic source ---
several modes (PP, PS, SP, and SS) can be labeled in the any source type can generate any combination of
composite records. wave modes.
(2) A wave mode may not be present in a record acquired
The soil column at the test site is largely composed of with a given source-receiver orientation which
interbeddings of marine clay deposited over a limestone- theoretically should give rise to that mode, but can
shale bedrock. Based on borehole information, the P-wave appear unexpectedly in a record acquired with a given
velocities vary from 930 m/s at the surface to 1,400 m/s at a source-receiver orientation which theoretically should
depth of 22 m with bedrock velocity of 3,500 m/s. not give rise to that mode.
Whereas, the S-wave velocities are 270 m/s at the surface, (3) The combination of wave modes captured by a
decrease to nearly 100 m/s at a depth of 10 m, then increase specific source-receiver orientation depends on the
to 200 m/s at a depth of 22 m with bedrock shear-wave Vp/Vs ratio.
velocity nearly 2,100 m/s. The Vp/Vs ratio varies from 9 to (4) These observations led to a realization that for a
1.7 within the soil column. complete representation of the wavefield propagating
The highest amplitude reflection of the top bedrock within the near-surface, we need to record
interface present at a depth of 22 m is observed at 260 ms multicomponent data.
for the SS-wave; its amplitude is highest in the V-geophone
gather; whereas in the H1-geophone gather, its peak
amplitude decreases and merges with a high and broad peak Acknowledgements:
which is likely associated with a mix of PS-SP and
refraction energy. These two peaks also merge in the H2- We thank Kevin Brewer, Timothy Cartwright and Matthew
geophone gather with little possibility to separate the SS Griffiths for their help in the field data acquisition and
mode from other modes. Even with the source orientation Gilles Bellefleur for providing corrections to the
in H2, the highest energy comes back on the V geophone. manuscript. This research has been supported by the GSC
The semblance analysis of the V-geophone gather exhibits Groundwater Geoscience activities.
detailed patterns of reflections from surface down to
© 2017 SEG Page 5163

Figure 2. 9-C seismic data recorded using a hand-held hammer and a receiver cable with 48 3-C (V: vertical, H1: inline
horizontal, and H2: crossline horizontal) 28-Hz geophones at 0.75-m intervals. We recorded three triplets of shot records (each
represented by the columns) with the impact source in vertical (V), inline horizontal (H1), and crossline horizontal (H2)
orientations. The receiver spread length is 35.75 m and the near-offset is 0.75 m. The labeling convention is as follows: the
first index represents the source component and the second index represents the geophone component. As an example, record
H1V was recorded using an inline horizontal source (H1) and a vertical source (V).
© 2017 SEG Page 5164

There is no pure P- or S-wave source
Figure 3. Left column: Composite records constructed by vertically summing the records shown in each row of Figure 2 with
three different source orientations (V, H1, H2), but with one geophone orientation. Center column: the semblance spectra of
the records shown in the left column. Based on velocities, several modes (PP, PS, SP, and SS) can be labeled in these records.
© 2017 SEG Page 5165

EDITED REFERENCES
REFERENCES
Hardage, R., 2015, Simplifying and lowering the cost of S-wave reflection seismology through use of P-
wave sources: SEG-GSH Webinar.
Hardage, R. and D. Wagner, 2014, 2-D modeling of direct-S and direct-P wavefields - part1 - modeling
principles and examples: EGL Sponsor Report, available at the private Members Area of the EGL
Website.
Miller, G. F., and H. Pursey, 1953, The field and radiation impedance of mechanical radiators on the free
surface of a semi-infinite isotropic solid: Proceedings of the Royal Society A: Mathematical,
Physical and Engineering Sciences, 223, 521–541, https://doi.org/10.1098/rspa.1954.0134.
Pugin, J.-M. A., S. E. Pullan, and J. A. Hunter, 2009, Multicomponent high-resolution seismic reflection
profiling: The Leading Edge, 28, 1248–1261, https://doi.org/10.1190/1.3249782.
© 2017 SEG Page 5166


Using finite difference modeling to understand direct-SV illumination produced by P sources

Donald Wagner and Bob Hardage*, Bureau of Economic Geology, The University of Texas at Austin
Summary
Numerical modeling of seismic responses created by specific geological conditions provides valuable information about the
fundamental physics of wavefield illumination, reflection wavelet character, wave-mode interactions, effects of inter-bed
multiples, and other elements of seismic reflection seismology. This presentation summarizes how finite-difference modeling can
be particularly valuable to researchers who are developing technology that allows full-mode, direct-SV reflection seismology to
be practiced with P-wave sources.
Introduction
Finite-difference modeling has been a critical asset in our efforts to establish the principles of direct-SV illumination and imaging
of targeted geology with P sources. Examples of our use of finite-difference modeling have been: (1) confirming the geometrical
shapes and relative amplitudes of direct-P and direct-SV radiation patterns produced by P (vertical displacement) and S
(horizontal displacement) sources, (2) determining how the responses of direct-P and direct-SV illuminating wavefields produced
by a P source are distributed between vertical and horizontal geophones, and (3) creating common-source gathers recorded by
vertical and horizontal geophones for a P source and identifying reflections created by downgoing, direct-SV, illuminating
wavefields in these trace gathers. In each model study, we created a 2D (X, Z) earth model consisting of small ∆X, ∆Z cells
(typically 6-m by 6-m in size). This grid of cells extended long distances left, right, and vertically from a source station, and each
grid cell was populated with Vp, Vs, and density values that were either assumed, or were provided by local log data. Vertical
and horizontal sensors were then positioned around a source station at small offset increments of typically 12-m at model
coordinates where we needed geophones to be to answer a specific technical question (e.g., as a long surface-based receiver
profile, or as a long vertical array of receivers in the subsurface, or as a semi-circular subsurface deployment of receivers
centered beneath a source station, etc.). The numerical calculations of direct-P and direct-SV illumination and reflection in this
large grid of finite-difference cells required access to large computer clusters and fast mass-storage devices, which are resources
that are not available to some investigators. Examples that illustrate the value of these finite-difference modeling efforts will be
shown and discussed.
Mathematical background
Any modeling code used to study wavefield propagation physics of direct-P and direct-SV modes produced by a P source must
correctly describe how vertical-displacement and horizontal-displacement sources generate P and SV radiation. For example,
some codes force the illuminating wavefield produced by a P source to be only a direct-P mode, which is a serious error. The
code used in our modeling accomplishes the fundamental requirement of allowing vertical-displacement and horizontal-
displacement sources to produce both direct-P and direct-SV modes by applying the following conditions:
(1) u = φ + x ψ
(2) uPi(x,t) = (1/4πρα2) γiγj (1/r)X0(t – r/α)
(3) uSi(x,t) = (1/4πρβ2) (δij – γiγj) (1/r)X0(t –r/β)
where u = displacement field, φ= P-wave potential, ψ = S-wave potential, φ= P-wave component of u, x ψ = S-wave
component of u, γjX0 = point source acting in direction j, uPi = displacement parallel to direction from source, uSi = displacement
normal to direction from source, γ = direction cosine, δ = Kronecker delta, α = P-wave velocity, β = S-wave velocity, and ρ =
density.
Calculating P and S radiation patterns
Examples of numerical calculations of direct-P and direct-SV illumination produced by a P source are illustrated in Figure 1.
Figure 1a shows the semicircular deployment of subsurface sensors that recorded direct-P and direct-SV wavelets radiating away
from a surface-based P source in take-off angles ranging from horizontal to vertical. Figures 1b and 1c show the character of the
direct-P and direct-SV radiation patterns produced by a P source for various Vp/Vs conditions in the medium local to a source
station. Fundamental principles established by these model data are: (1) P sources produce more direct-SV energy than direct-P
energy, (2) the amount by which direct-SV energy exceeds direct-P energy increases as the Vp/Vs ratio increases at the source
station, and (3) the geometrical shapes of P radiation patterns are fundamentally different from the geometrical shapes of SV
radiation patterns. These P and SV radiation patterns are correct only for a homogenous medium. Examples of direct-P and
direct-SV radiation patterns produced by a P source when irregularities such as tree roots, lateral variations in soil texture, or
breaks in surface flatness, are near a source station will be shown and discussed.
1

© 2017 SEG Page 5167

Constructing synthetic common-source trace gathers
Probably the most informative use of finite-difference modeling in our research has been the construction of synthetic shot
gathers to assist the processing and interpretation of SV-P data produced by P sources. In each of these modeling efforts, we used
a dipole sonic log and a formation density log acquired in a local well to assign Vp, Vs, and rho values to a vertical stack of
finite-difference cells at the well, and then extended these 1D velocity and density profiles laterally across the entire 2D model
space. We ensured that no reflections from model boundaries were included in calculated shot gathers by extending grid cells at
least 9000-m in all directions from each source station.
Value of calculated common-source gathers in extracting SV-P reflections
Principal uses of P-source synthetic shot gathers are to illustrate how direct-P and direct-SV reflections overlap and interfere in
vertical-geophone data, and to test procedures for extracting SV-P reflections from their competing P-P reflection background.
Conversely, these models also allow data processors to ensure that P-P reflections are optimally segregated from their interfering
SV-P reflection background. An example of a common-source gather of vertical-geophone traces used in this manner is
illustrated in Figure 2. Velocity logs that provided the Vp and Vs velocities used to populate the finite-difference cells are shown
in Figure 2a. The calculated shot gather in Figure 2b has been muted to eliminate extreme-offset data, and a log-based, SV-P
normal-moveout (NMO) correction has been applied to the data. Because of this NMO application, any SV-P reflections created
by the targeted geology should be flat across the trace gather. There is only faint evidence of flat events in the data in Figure 2b.
The SV-P NMO correction also causes P-P reflections to be over-corrected and to curve upward, and SV-SV reflections to be
under-corrected and curve downward. In real trace gathers, SV-SV reflections are rarely found in vertical-geophone data. SV-SV
reflections occur in these synthetic data because the depth interval between the top-most, log-defined, Vs cell and the earth
surface is filled with a constant Vs value that is faster than the unknown, real-earth Vs values in the near-surface layers. This “too
fast” Vs velocity in the near-surface interval of the model does not allow upgoing SV raypaths to be vertical at receiver stations,
which results in vertical components of upgoing SV-SV reflections that activate vertical geophones.
Figure 2c shows the trace gather in Figure 2b after up-curving events are attenuated. This procedure uncovers several flat SV-P
reflections that were hidden by P-P reflections in Figure 2b. Figure 2d shows the data in Figure 2c after down-curving events are
attenuated, and more flat SV-P reflections are discovered when interfering SV-SV reflections are removed. These data-
processing steps show how synthetic shot gathers allow data processors to test various procedures for attenuating events that
interfere with SV-P reflections. In our research, we have developed a preference for following the data-processing procedure
illustrated here; i.e., apply SV-P NMO to shot gathers and then attenuate up-curving and down-curving events with algorithms
that are not overly harsh and thus force a biased outcome.
Value of calculated common-source gathers for identifying SV-P reflections
Data processors have favorite techniques for generating various types of trace gathers. Regardless of the method used, it is
invaluable to have an appropriate synthetic trace gather, in which log-based SV-P reflections can be identified, to compare with
the real-data trace gather. The example shown in Figure 3 illustrates how a legacy CDP trace gather (Figure 3a) that became
available was quickly examined with an existing synthetic shot gather (Figure 3b) in which log-based SV-P reflections were
isolated as flat events. Even though the wavelet used in the synthetic model had a signal bandwidth much lower than the wavelet
embedded in the real-data trace gather, the comparison quickly verified that the down-curving events in the CDP trace gather
were SV-P reflections and caused a refocused data-processing effort to be initiated to create an SV-P image from these vertical-
geophone data.
Conclusions
There is nothing fundamentally new in the general theme emphasized in this presentation; i.e., that log-based modeling is
valuable in processing and interpreting seismic data. The “new” aspect of the examples that are shown is that synthetic data are
particularly important in developing technologies that will allow SV-P reflection seismology to be done using vertical-geophone
data generated by P sources. No examples are yet in the geophysical literature that focus on SV-P illumination and reflectivity
associated with P sources. We are fortunate to have access to computational resources that allow us to use finite-difference
modeling to create the synthetic data we use. Others may not have access to such resources, or to appropriate code that creates
realistic direct-P and direct-SV illuminating wavefields. Thus new fields of investigation are now open to develop modeling
codes that will explain SV-P physics associated with P sources, and that assist the development of SV-P data processing and
interpretation. It is essential that whatever modeling procedures will be utilized illustrate how P-P, SV-P, and SV-SV reflections
produced by P sources interfere with each other, and then allow data-processing algorithms to be applied to the digital model data
to test methods for extracting high-quality SV-P reflections.
2

© 2017 SEG Page 5168

Figure 1. Calculation of direct-P and direct-SV radiation by a P source. (a) Source-receiver geometry used in the calculations. (b)
Direct-P and direct-SV radiation patterns when the source station is located in a medium having a Vp/Vs of 2, and (c) in a
medium having a Vp/Vs value of 4.
Figure 2. (a) Velocity logs used to populate grid cells. (b) Synthetic common-shot trace gather created by finite-difference
modeling with offset mutes and a log-based SV-P NMO applied. (c) Data after up-curving P-P reflections are attenuated. (d) Data
after down-curving SV-SV reflections are attenuated. The final result shows an impressive family of SV-P reflections.
3

© 2017 SEG Page 5169

Figure 3. Using a synthetic shot gather, made with a low-frequency wavelet in which SV-P reflections are isolated and flattened,
to verify that down-curving events in a legacy CDP gather of migrated P-P traces are SV-P reflections.
4

© 2017 SEG Page 5170
EDITED REFERENCES
REFERENCES
No references.
© 2017 SEG Page 5171

Gas reservoir characterization using SV-P converted wave mode, a case study from Western
China
Yandong Li*, Research Institute of Petroleum E&D, PetroChina, Daxing Wang, Songqun Shi, and Xiaojie Cui,
Research Institute of Petroleum E&D, Changqing Oil field Company, PetroChina
Summary
Geology and reservoir description
The capability of “seeing though” the gas-charged zones is
one of the most important benefit of converted S-wave from The study area is located in the Sulige Gas Field, the largest
the very beginning of multicomponent seismic study. onshore gas field in China. It is located in the western part
Usually the converted S-wave used in gas-charged zone of the Yishan slope in the Ordos Basin. The source rock is
imaging is the P-SV wave mode. SV-P is an emerging coal and mudstone of Benxi Formation of the Upper
converted-wave technology proposed by the Bureau of Carboniferous, Taiyuan and Shanxi Formations of the
Economic Geology at the University of Texas at Austin. The Lower Permian. The main gas bearing interval is the He 8
equivalence of the imaging capability between SV-P and P- Member of the Lower Shihezi Formation and the Shan 1
SV wave modes has been established both by theoretical Member of the Shanxi Formation of the Lower Permian
analysis and by real data imaging applications. This paper (Zhu et al., 2009). The fluvial and delta deposits are
presents an example of gas reservoir characterization using developed in the Lower Shihezi Formation on the gentle
SV-P wave mode extracted from legacy P-wave data Yishan slope (1) of the Ordos Basin. The thickness of the
acquired by vertical geophones using explosive sources. We Lower Shihezi formation is between 220 – 280 m and the He
compared the SV-P image and the P-P image and found that 8 Member has a thickness between 50-100 m and lays at a
SV-P image well delineates the top and base of the prolific depth of approximately 3000 m in study area. The He 8
gas reservoir whereas these interfaces are almost invisible on Member is proved to be more prolific than Shan 1 Member
P-P image. Synthetic SV-P and P-P seismograms using well and is the focus of the SV-P based gas reservoir
log data confirmed this contrast. These results demonstrate characterization study in this paper.
that SV-P could be a valuable wave mode in reservoir
characterizations. Description of the seismic dataset
Introduction The 3D seismic survey was acquired using vertical

geophones and conventional P-wave explosive sources. The
SV-P images should be as informative and as valuable as P- receiver line and shot line spacing are both 280 m. The
SV images for seismic interpretation purposes (Hardage et receiver interval and source interval are both 20 m. The bin
al., 2014). The difference between SV-P and P-SV wave size for the conventional P-wave based processing is 20 m
mode is that the former is acquired by conventional vertical ×10 m, with inline bin size being 20 m and the cross-line
geophones and the latter can only be acquired by 3C bin size being 10 m. The average fold number for the P-P
horizontal geophones. One particular attraction of the SV-P reflections at this bin size is about 130 at the target level. The
technology is that valuable S-mode images and attributes surface at the study area is mainly covered by desert, loose
could be extracted from legacy P-wave data without soil, alkali beach and shrubs which have severe lateral
incurring new seismic data acquisition cost. This peculiar variations in P-wave and S-wave velocities, posing big
feature of SV-P technology motivated us to initiate an SV-P challenge to static correction and noise suppression.
processing for the legacy P-wave data acquired in 2013 at
the Ordos Basin, Western China. The dataset is a typical P- Verification of the existence of SV-P reflection
wave dataset in that it was generated by explosive sources
and acquired by single-component vertical geophones. After The first step in SV-P processing for the legacy P-wave data
preliminary feasibility study on the SV-P processing for this is to establish the existence of reliable SV-P reflections in
legacy P-wave dataset, we conducted SV-P oriented the conventional P-wave data acquired by vertical
processing and the results show that SV-P image is able to geophones. A typical shot gather is shown in Figure 1a, in
clearly delineate the top and base of the prolific gas reservoir which the two marked P-P reflections from two coal seams
formation whereas these interfaces are almost invisible in the are obvious. Since there are no dipole sonic logs within the
P-P image due to the reduced P-impedance contrast between survey, we use a nearby well with similar geology to build
the sealing shale and the gas-saturated sand formation. The rock physics model for S-wave velocity estimation for the
results are exciting and give us great confidence in the calibration well A1 in the study area. Based on the estimated
successful applications of SV-P technology in reservoir S-wave curves and measured good quality P-wave and
characterization. density logs, we performed elastic wave-equation and
© 2017 SEG Page 5172

Gas reservoir characterization using SV-P
acoustic wave-equation based forward modeling using the static corrections, we finally produced a subset of 3D SV-P
geometry of the real shot gather in Figure 1a. The modeled image volumes. An Inline profile through the SV-P image
gathers are shown in Figures 1b (elastic modeling) and 1c volume intersecting the calibration well A1 is displayed in
(acoustic modeling), respectively. Since there are no Figure 2b.
converted waves in Figure 1c, we can identify the SV-P
reflections in Figures 1a and 1b. The reflection times of SV- To prove the correctness of the SV-P image in Figure 2, we
P and P-P reflections for the two coal seams agree with the generate synthetic seismogram for the P-P and SV-P wave
relationship between P-P and SV-P given in Tessemer and modes at the calibration well A1 and then perform well-
Behle (1988). The good consistency between the modeled seismic tie. Figure 3a displays the synthetic SV-P
shot gather and the real shot gather verified the existence of seismogram (red) and the corresponding real SV-P image
reliable SV-P reflections in the real seismic data acquired by traces (black) and Figure 3b shows the synthetic P-P
vertical geophones. seismogram (red) and the corresponding P-P image trace
(black). The relatively good correlations between the
Comparison of SV-P image with P-P image synthetic and real seismograms illustrate the correctness of
P-P and the SV-P imaging results. There are conspicuous
The P-P imaging is not a too difficult job in the study area differences between the SV-P and P-P synthetic and image
except for the static corrections caused by severe near traces, with SV-P having much richer reflections than their
surface variations. Anyway, the commercial seismic P-P counterparts.
processing software packages are able to tackle these
problems and produce a reasonably good result. An Inline To further compare the P-P and SV-P image in a larger range,
seismic profile intersecting the calibration well A1 from the we display a longer P-P and SV-P profile (approximately 5
3D P-P image volume is displayed in Figure 2a. For SV-P km) in Figure 4. The position of the seismic profile in Figure
imaging, it is not that easy. Conversely, it is still quite 4 is marked by the black line C-C’ in Figure 5, which shows
challenging to produce a good SV-P image for real legacy P- the distribution of gas reservoir produced by conventional P-
wave seismic data. After a great deal of tough and wave based reservoir characterization methods. The warm
intermingled processing steps such as P-P and SV-P color in Figure 5 indicates thick and porous gas reservoir and
reflection segregation, SV-P oriented surface-consistent the cool color represents thin low-quality reservoir. There
deconvolution and amplitude compensation, SV-P oriented are four calibration wells named A1-A4 in Figure 5. Among
Figure 1: (a) Real shot gather and modeled shot gather using elastic wave equation (b) and acoustic wave equation (c). The two strong
P-P reflections are from two coal seams, which constituent excellent marker horizons for registering P-P and SV-P reflections. The P-
P multiples are markedly different from the SV-P reflections which are weak at near offsets.
© 2017 SEG Page 5173

Figure 2: Inline profile (south to north along line C-C’ in Figure 5) intersecting the calibration well A1 through the 3D P-P image volume
(a) and the SV-P image volumne (b). H1 and H2 are two marker horizons. The top and base of the drilled gas reservoir have been clearly
delineated in the SV-P image but is almost invisible on the P-P image.
these four wells, A1-A3 (marked by solid black stars) are decreased impedance contrast between sealing shale and the
prolific gas wells while A4 (marked by hollow white circle) gas-saturated sandstone. This will cause weak P-P
being a dry hole. reflections at the top and base of the gas-saturated sand
reservoirs. In contrast, the S-waves are insensitive to fluids
and the gas-saturated sand reservoirs still possess higher S-
impedance than the sealing shale. Therefore, strong
reflections are produced at the top and base of the gas-
saturated sand reservoir in the SV-P image.
Discussions
Our industry has struggled for more than 50 years to find

cost-effective and practical S-wave applications. However,
the S-wave still has not become a significant business driver
(Gaiser and Strudley, 2005). The SV-P image in this paper
is far-from being perfect. However, such a sub-optimal SV-
P image already exhibits great advantage in gas reservoir
characterization. Therefore, we should rebuild our
Figure 5: Gas distribution delineated by P-P based methods. Warm confidence in 3C seismic technology to pursue its added
color indicates thick porous gas layer and dark color represents value to P-waves. We are of the opinion that P-wave and S-
thinner ones. Line C-C’ marks the position of the profile in Figure
wave are not enemies; instead they are intimately connected
4. A1-A3 are three prolific gas wells and A4 is a dry hole.
from the very beginning of their generation by the same
There are marked differences between the SV-P image and source. They should help each other to gain a better
P-P image in Figure 4, especially in the proved prolific gas understanding of our mother earth.
reservoir zone marked by the blue rectangles. In these gas-
charge zones, P-P image is smeared while the SV-P image Conclusions
clearly delineates the top and base of the gas reservoir. The
physical mechanism behind this phenomenon is well-known Strong and reliable SV-P reflections can be extracted from
to experienced geophysicists. That is gas-saturation causes legacy P-wave data acquired by single vertical geophones
considerable decrease in P-impedance, resulting in a using conventional explosive P-wave sources. These SV-P
reflections could be imaged and produce a better
© 2017 SEG Page 5174

characterization of the gas reservoir than their P-P

counterparts. Further studies are needed to additionally

improve the quality of SV-P image and to extend its
application to geologically more complex reservoirs.
Acknowledgments
This work is partially-supported by the State Key Science

and Technology Project on Marine Carbonate Reservoir
Characterization, China (2016ZX05004-003) and Chinese
National Natural Science Foundation project (41504110).
Figure 3: (a) SV-P image traces (black) around the calibration well A1 and the corresponding synthetic seismogram (red), (b) P-P image
traces (black) around the calibration well A1 and the corresponding synthetic seismogram (red). H1 and H2 are two marker horizons.
Much richer reflections exist in the SV-P synthetic and real image traces than their P-P counterparts.
Figure 4: Comparison of P-P and SV-P image. The SV-P image clearly delineates the top and base of the gas reservoir,
whereas these interfaces are almost invisible on P-P image. The blue rectangles mark the proved gas-charged zones.
© 2017 SEG Page 5175

EDITED REFERENCES
REFERENCES
Gaiser, J., and A. Strudley, 2005, Acquisition and applications of multicomponent vector wavefields: Are
they practical? First Break, 23, 61–67, http://doi.org/10.3997/1365-2397.2005012.
interpreters: Interpretation, 2, SE17–SE27, http://doi.org/10.1190/INT-2013-0096.1.
Tessemer, G., and A. Behle, 1988, Common reflection point data-stacking technique for converted waves:
Geophysical Prospecting, 36, 671–688, http://doi.org/10.1111/j.1365-2478.1988.tb02186.x.
Zhu, X. M., C. L. Liu, D. K., Zhong, and X. F., Han, 2009. Diagenesis and their succession of gas-bearing
and non-gas-bearing reservoirs in the Sulige Gas Field of Ordos Basin, China: ACTA Geologica
Sinica, 83, 1202–1213, http://doi.org/10.1111/j.1755-6724.2009.00087.x.
© 2017 SEG Page 5176

SV-P imaging compared to P-SV imaging: analysis of statics and velocities required to create an
SV-P image
Bruce Karr*, FairfieldNodal, Houston, Texas
Summary
Standard 3D seismic data recorded into traditional vertical geophones can produce an SV-P converted-wave data volume.
However, there is a challenge in resolving both the statics and velocities while separating out the SV-P image from the P-wave
portion of the data. The primary difficulty is in resolving the velocity and statics associated with the signal of interest, SV-P,
which is buried in the slower velocity and deeper time portion of the seismic P-wave record. In other words when looking at a
gather in offset order and moved out with a P-wave velocity, you will often see many under moved out events and more of those
events with increasing time. This is the portion of the gather that has always been referred to as containing noise, coherent noise,
multiples, and mode-converted responses. This presentation looks at a real-data example which took advantage of using the
traditional P-SV processed converted-wave-stack velocities and statics as the basis for looking for the SV-P equivalent stack. In
particular, if we know the velocity field and the final P-SV stack from traditional converted wave data, can we successful resolve
statics and velocities to create the SV-P image? This data example shows it can be done.
Introduction
The observance of converted and even shear direct arrivals on downhole VSP data from a P-wave source has been documented
(Hardage and Wagner, 2014a, 2014b; Hardage et al., 2014; Li and Hardage, 2015). This presentation example involves high
quality traditional 3D surface seismic P-to-P and P-SV converted-wave data volumes. Therefore it proved to be a good survey to
search for the SV-P image in the P-wave record and compare that image with a reliable P-SV image. The relatively high quality
reflections show there is very good signal in the P-wave data at the gathers and stacked stages. P-SV data recorded by the
horizontal geophones also shows good reflectivity and provided a fairly easy registration between the P-P and P-SV images. The
first challenge proved to be the receiver statics on the converted-wave P-SV data. The receiver statics turned out to be a
magnitude of 10 greater than the P-wave receiver statics. The second challenge was looking for the SV-P velocity field in the
traditional P-wave records. Approaches for estimating S-static corrections required for converted-mode data, and particularly for
SV-P data, will be discussed in this presentation. Velocity analyses are also a key data-processing step when making seismic
images with any wave mode, and are particularly important when data-processing objectives are to make images with converted-
mode data. Concepts that have proven to be useful in determining optimal stacking velocities for SV-P data will be discussed in
this presentation.
Comparison of the P-wave image to the converted wave P-SV image
The presentation starts with a look at the high quality P-wave stacked image and compares it with the traditional P-SV converted
wave image. Next, I review the P-wave gather and highlight the under-moved-out reflectors that are believed to be some of the
mode converted events that we are actually looking for to create the SV-P images (Figure 1). In conjunction with the gather, I
look at a standard P-wave constant velocity panel to estimate where mode- converted wave images would reside in the P-wave
recorded data volume (Figure 2).
S statics for SV-P imaging
The map in Figure 3 shows near-surface receiver S static values from the P-SV data. The color bar in the figure indicates the
magnitudes of the S-static corrections that exist across the image space. The data acquired in this seismic survey involved 3C
geophones and two types of P sources: buried explosives and vertical vibrators. The static corrections shown on this map were
determined from analysis of P-SV data acquired at receiver stations with horizontal sensors. Note the large variations in the S-
static estimations, which range from +30 ms to -80 ms across the image space, and the short lateral distances over which large
variations in S statics can occur. In order to process SV-P data, this S-static behavior, which is based on horizontal-sensor
responses, has to be translated into S-static corrections that are appropriate for applying at P source stations, not at receiver
stations.
An example profile through the processed 3D P-SV data volume is displayed in Figure 4. The image quality is excellent, which
confirms that S-statics were accurately estimated at horizontal-receiver stations. The corresponding SV-P profile, which required
that different S-statics be applied at the P-source stations (not at the receiver stations), and additional trim statics is shown as
Figure 5. Comparison of Figures 4 and 5 confirms that accurate source plus trim statics were required to create very similar stack
images. At this data-processing stage, it is obvious that the frequency bandwidth of the SV-P reflection signal (Figure 5) is less
than the frequency bandwidth of the P-SV reflection signal (Figure 4); however, the image detail is remarkably similar even at
the lower frequency response.
© 2017 SEG Page 5177

The will be logic and methodology that allowed accurate S statics to be determined at the P-source stations involved in this
seismic program illustrated and discussed in the presentation.
Velocity analysis for SV-P data

The final section of the presentation will look at examples of SV-P velocity analysis with comparisons to the P-SV velocity
results. Part of the discussion will illustrate how stacking velocities needed for converted-mode imaging (SV-P and P-SV data)
can be determined by examination of constant-velocity stacks of single-mode data (P-P and S-S modes) across the same image
(Figure 2 again). This discussion about the interplay of stacking velocities for converted-mode data and single-mode data will be
an important piece of this presentation and they, too, are not easily resolved.
Conclusions
SV-P data are contained in traditional P-P mode traces. Static estimations and velocity analyses needed for imaging SV-P data
can be challenging, but are not an impossible task. The experience that has been learned in several projects will be shared to aid
others to make progress toward creating SV-P images from data generated by P sources and recorded by vertical geophones. One
strong take away for continuing the development of SV-P processing is to continue to leverage data sets that have P-SV data to
compare against. Another strong take away would be to find areas where shear wave statics are much smaller in magnitude than
the statics dealt with in this analysis. One basin that comes to mind as having modest static corrections is the Midland basin.
Figure 1. Traditional processed P-wave gather. Processing includes pre-processing wavelet correction, refraction statics, 1st pass
surface consistent residual statics, 1st pass of velocities, and a FX decon filter. Green line is the approximate mute to be applied
based on NMO. Under-moved-out events represent possible mode-converted events.
© 2017 SEG Page 5178

Figure 2. This figure is a P-wave constant-velocity panel based on midpoint binning. Red represents 1st pass P-wave velocity
picks. Green projects where the 1st pass P-SV or SV-P velocity would be if a constant Vp/Vs=1.8 were the case. Orange boxes
represent possible SV-P packages of geology and where they would approximately reside on the P-wave constant velocity panel.
If Vp/Vs is 1.7, it would shift the green marks more into the orange boxes. 1.7 to 1.8 are very reasonable Vp/Vs values for this
region. The binning, however, is still mid-point and not ACP binned like traditional converted wave data would be.
Figure 3. Area spanning near-surface conditions where S-static corrections change rapidly with distance and exhibit large
variations in magnitude. However, the statics do follow patterns that are probably related to the near surface weathering geology.
© 2017 SEG Page 5179

Figure 4 top, Figure 5 bottom.

Comparison of traditional post-stacked migrated data of P-SV (Figure 4) and SV-P mode-converted data from the P-wave source
into the vertical geophone (Figure 5).
© 2017 SEG Page 5180

EDITED REFERENCES
REFERENCES
Hardage, B.A., and D. Wagner, 2014a, Generating direct-S modes with simple, low-cost, widely available
seismic sources: Interpretation, 2, SE1–SE15, https://doi.org/10.1190/INT-2013-0095.1.
Hardage. B.A, and D, Wagner, 2014b, S-S imaging with vertical-force sources: Interpretation, 2, SE29–
SE38, https://doi.org/10.1190/INT-2013-0097.1.
Hardage, B.A., D. Sava, and D. Wagner, 2014, SV-P: An ignored seismic mode that has great value for
interpreters: Interpretation, 2, SE17–SE27, https://doi.org/10.1190/INT-2013-0096.1.
Li, Y., and B.A. Hardage, 2015, SV-P extraction and imaging for far-offset vertical seismic profile data:
Interpretation, 3, SW27–SW35, https://doi.org/10.1190/INT-2015-0002.1.
© 2017 SEG Page 5181

Improved reservoir delineation by using SV-P seismic data in Wellington field, Kansas
Menal Gupta* and Bob Hardage, Bureau of Economic Geology, The University of Texas at Austin
Summary mode-converted into up-going P-waves that are recorded on

A traditional converted-wave (P-SV) seismic acquisition vertical geophones. This wave physics eliminates the need
requires 3C geophones and thus more cost compared to for using a 3-component receiver to record an up-going
conventional P-P data. However, some companies justify wave-field. Principles and algorithms of CCP processing
this added cost because converted-wave data provide an originally developed for processing P-SV data can be easily
independent set of shear-wave measurements, which applied to SV-P data processing.
increase the reliability of subsurface property estimation.
This study investigates SV-P data generated by P-wave Here, we evaluate the efficacy of the SV-P mode to
sources and recorded by vertical geophones as a cost- understand subsurface properties and compare this estimate
effective alternative to traditional P-SV data. To evaluate the with the estimates of properties obtained from P-P data
efficacy of the SV-P mode, multi-component seismic data alone. Results show that P-P data provide excellent P-
from Wellington Field, Kansas were interpreted. P-P AVO impedance (IP) information, but that it is essential to use
gathers and stacked SV-P seismic data were jointly inverted high-quality well log data and P-wave AVO gradients to
to estimate elastic properties, and compared with the obtain S-impedance (IS). Results also show that use of SV-
estimates obtained from inversion of P-P AVO gathers data. P data offsets this problem because these data provide an
A significant improvement in S-impedance estimates is independent S-wave measurement and more reliable
observed when SV-P data are used in addition to P-P AVO information about IS behavior.
gather data. In the Arbuckle interval, which is being
considered for CO2 injection, use of SV-P data clearly Study area and dataset
demarcates Middle Arbuckle baffle zone and the Lower Wellington Field is located in Sumner County, Kansas and
Arbuckle injection zone. These observations, while has been producing from cherty, dolomitic reservoirs of
consistent with other well-based geological evidence, are Mississippian age. In addition to Mississippian reservoirs,
absent on P-P only inversion results. Moreover, comparing stratigraphic units that are of immediate interest to this study
the SV-P image obtained from vertical-vibrator data and the are the Lower Ordovician carbonates that are collectively
SV-P image obtained from horizontal-vibrator data called the Arbuckle Group. The Lower Arbuckle is believed
establishes that both images are essentially equivalent, to be heavily dolomitized and fractured and is being
except for the important distinction that former contains considered for CO2 storage (Watney and Rush, 2012).
valuable higher frequencies than latter. Natural fractures have been interpreted in the Lower
Arbuckle with well-logs, such as micro-resistivity and
Introduction dipole sonic logs (Watney and Rush, 2012), and with multi-
P-SV data have been the most popular data choice among component seismic data (Gupta et. al, 2016) acquired in
proponents of multi-component seismic technology. Wellington Field. Seismic-scale fractures are assumed to
Reasons for this popularity are that P-SV data: (1) provide facilitate fluid flow in the Lower Arbuckle.
an independent set of shear-wave measurements with
applications in estimating subsurface density, and To better understand the role of fractures and heterogeneities
anisotropy, (2) can be generated by a conventional vertical- in the CO2 injection interval and overlying sealing
force source, and (3) are more practical in terms of data formations, a 9C-2D seismic survey was acquired across
volume and acquisition cost, compared to S-S mode data. Wellington Field. Data from Line 1 has been used in this
Despite these advantages, there is still reluctance in the study. Horizontal vibrators made six linear sweeps (20 s
geophysical community towards incorporating P-SV mode long) with frequency ranging from 4Hz to 50Hz to make
into routine interpretation projects. Reasons are: (1) maximum utilization of the lower end of the frequency
acquiring P-SV data requires 3-component receivers, which spectrum. Vertical vibrators on the other hand, made two
adds to cost and complexity of data acquisition, and (2) P- non-linear sweeps (40 s long) ranging from frequencies 6Hz
SV seismic data are sometimes noisy, because of the to 150Hz with a taper of 3db/octave. The 2D
difficulty of coupling horizontal geophones to the earth and multicomponent seismic data were processed at a
tendency of horizontal receiver to be more noise sensitive. commercial data-processing shop where they were
segregated into various wave modes including direct P (P-P)
SV-P data generated from vertical-force sources (Gaiser and and direct S (S-S), converted-P (SV-P), and converted-S (P-
Verm, 2012; and Hardage and Wagner, 2014a, b, and Li and SV). The SV-P wave mode generated by the vertical vibrator
Hardage 2015) offers an effective solution to offset these sources was also processed.
challenges. SV-P data makes use of the down-going shear Two wells, KGS 1-28 and KGS 1-32, are located directly on
wave-field generated directly at the source, which then is the 9C-2D seismic survey lines. Both wells have a full suite
© 2017 SEG Page 5182

Double click here to type your header
of well-logs including spectral gamma ray, neutron density, of P-SV section (Figure 2b) deteriorates NE of well 1-28 due
resistivity, cross-dipole sonic, nuclear magnetic resonance to poor S-wave static corrections.
(NMR) and micro-resistivity imaging. In addition to well-

logs, lithology data from cuttings and information about rate
of drill bit penetration (ROP) are also available as shown in
Figure 1. Greater abundance of fractures and vugs in the
Arbuckle Group has been reported at all scales based on
well-log and core data. These vugs and fractures have been
found to govern overall permeability in the Lower Arbuckle
(Gupta et al., 2014 and 2015; Ohl and Raef, 2014; Watney
and Rush, 2012; Franseen et al., 2003; and Byrnes et al.,
1999).
Figure 2: P-P (a) and P-SV (b) stacked seismic section showing the
Figure 1: Well log suite acquired in well 1-32. The area shaded key interpreted seismic horizons on Line-1. P-P seismic has better
yellow in the Gamma Ray (GR) track corresponds to the carbonate signal-to-noise ratio compared to the P-SV seismic section. Polarity
intervals in the well. Shaly intervals are shaded gray. S-wave on the P-SV seismic section has been matched to P-P seismic to see
anisotropy is shown in the Aniso track. Rate of Penetration (ROP) event correspondence. Within the Arbuckle Group, high amplitude
and lithology estimates for the Arbuckle group are digitized from reflectors in the Lower Arbuckle injection zone (green highlighted
the mud-log of well 1-32. Middle Arbuckle (between markers A and zone) are observed on both seismic sections.
B) shows a higher GR and slower ROP compared to the Lower
Arbuckle (below marker B), suggesting the Middle Arbuckle is a To ascertain the fidelity of SV-P data generated by a vertical
tighter formation. This variation is not captured by the seismic vibrator, vertical-vibrator SV-P data were compared to SV-
velocities measurements made at well-log scale.
P data generated by a horizontal-vibrator, the source that is
the gold standard for generating down-going illuminating S-
Multicomponent seismic data interpretation
waves. Figure 3a and 3b show SV-P images generated by a
To compare and interpret P-P and various mode-converted
horizontal and vertical vibrator, respectively. The SV-P data
modes (P-SV and SV-P), stacked images were created from
shown in Figure 3b are derived with a vertical-vibrator
corrected gathers obtained from the data-processing
source and vertical geophone. As a result, this image is
contractor (FairfieldNodal) as shown in Figure 2 and 3. All
generated from the same source-receiver setup that is used
these images were depth registered (Gupta et. Al, 2015) and
to acquire conventional P-wave seismic data.
converted to P-P time. Seismic horizons corresponding to
key geological boundaries like Kansas City, Mississippian,
In addition to a reasonable correspondence between the two
Arbuckle and Precambrian have been interpreted on all sets
images, SV-P data generated by a vertical vibrator also
of seismic data. The Arbuckle Group, which is the primary
provide higher resolution compared to SV-P mode generated
zone of interest, stretches from the Arbuckle horizon (at 710
from a horizontal vibrator, particularly close to the two
ms) to base Arbuckle horizon (at 810 ms) in P-P time. The
calibration wells. This improved resolution happens because
onset of the baffle zone in the Middle Arbuckle is marked by
the sweep frequencies of the vertical vibrator (6-150 Hz) are
a high amplitude seismic reflection marked as horizon A (at
higher compared to the sweep frequencies of horizontal
740 ms), which can be seen on P-P (Figure 2a) and P-SV
vibrator (4-50 Hz). Thus, the SV-P image from the vertical
(Figure 2b) and SV-P images (Figure 3). Also, seismic
vibrator should have higher frequencies as can be seen in
reflectors corresponding to the Lower Arbuckle stretch from
Figure 4. Figures 3 and 4 demonstrate that SV-P data
horizon B (at 775 ms) to base Arbuckle, with the proposed
generated from a conventional P-wave source can be a viable
CO2 injection zone highlighted in green. The image quality
alternative to both the P-SV mode produced by a P-wave
© 2017 SEG Page 5183

source and SV-P seismic mode generated by S-wave Inversion of multicomponent seismic data
sources. After the structural interpretation of P-P, P-SV and SV-P
data is completed, a quantitative analysis of all data sets is

warranted to understand the value that mode-converted data
bring to subsurface property estimation. Two sets of
inversions (P-P AVO inversion and P-P gathers and stacked
SV-P joint inversion) were performed using Hampson
Russell Software and resulting P-impedance (IP) and S-
impedance (IS) were compared. AVO inversion of P-P data
is a widely accepted practice to estimate subsurface elastic
properties from conventional P-wave seismic data. The
second inversion used stacked SV-P seismic data generated
by a vertical vibrator in addition to P-P angle gathers. Figure
5 shows the pre-stack inversion results for the P-P AVO
trace gather data. P-P CDP gathers were converted to angle
gathers, which were then inverted to estimate IP and IS for
Line 1. IP results in Figure 5a shows good agreement with
the IP observed on the well logs, which are shown as a
narrow colored strip below each well label. Results show
high IP below the Kansas City reflector, where there is an
onset of thick limestone units. Similarly, the high-impedance
zone just below the Mississippi reflector coincides with the
onset of dolomite. The Middle Arbuckle shows high IP,
which indicates a possible change in litho-facies from the
Figure 3: A comparison of the SV-P image obtained from a Upper Arbuckle. Also, low IP values can be seen in the
horizontal vibrator (a) with the SV-P image obtained from a vertical Lower Arbuckle. Figure 5b shows the IS estimated from pre-
vibrator (b). Seismic reflections corresponding to Kansas City, stack P-P trace gathers. Within the Kansas City,
Mississippi, Arbuckle and Precambrian are shown on both Mississippian and Arbuckle intervals, IS seems to be in
converted-wave modes. Within the Arbuckle Group, top of the general agreement with measurements made on well-logs
Middle Arbuckle reflection (A) and top of the Lower Arbuckle
(vertical colored strips below well labels). The Middle
injection zone (highlighted in green) are consistent in both seismic
sections, demonstrating the efficacy of using a vertical vibrator to Arbuckle (A to B) shows high IS values. A smaller change
generate down-going S-waves. is observed in IS from Middle to Lower Arbuckle compared
to the inverted IP. Additionally, the proposed injection zone
is not well defined in Figure 5b.
Figure 4: A comparison of the frequency spectra of SV-P seismic

mode generated by P-wave source (red) and S-wave source (blue).
The SV-P mode generated by a vertical vibrator contains more high
frequency energy from 57 Hz to approximately 67 Hz. This increase
in frequency content is a result of using a P-wave source which
generates sweeps with broader frequency spectrum compared to an
S-wave source. These additional high frequencies help generate
higher resolution image of the subsurface enabling more reliable
interpretation. Figure 5: (a) P-impedance (IP) obtained from AVO inversion of P-
P gathers for Line-1 compared to IP computed from well-logs
© 2017 SEG Page 5184

(vertical colored strips below well labels). Middle Arbuckle interval inversion. Figure 6b shows IS for Line-1 compared to IS
(between A and B) shows high IP values, suggesting a possible computed from well-logs. IS estimated from P-P and SV-P
lithology change, consistent with presence of muddy crystalline joint inversion better resolves lithology variations associated
dolomite observed in core. This high impedance layer coincides
with Kansas City and Mississippi horizons, and within the
with the baffle zone known to be present in the Middle Arbuckle
(Watney and Rush, 2012). Lower Arbuckle injection zone (below Arbuckle Group, compared to IS estimated from P-P
horizon B) shows lower values of IP which coincides with the inversion (Figure 5b). The Lower Arbuckle injection zone
presence of fractured dolomitic limestone. (b) S-impedance (IS) (below B) shows low values of IS, suggesting variation in
obtained from AVO inversion of P-P gathers compared to IS rock properties as has been seen on IP. Inverted IS in the
computed from well-logs. Middle Arbuckle interval (between A and injection zone coincides remarkably with the zones of fast
B) shows high values of IS. Change in IS values from Middle to ROP and low GR values (refer to Figure 1), suggesting
Lower Arbuckle is considerable less compared to what was mechanically weaker rock. These variations are not as
observed on IP section. Lower Arbuckle injection zone is not clearly
evident on IS obtained from P-P AVO inversion results
visible on IS image.
(Figure 5b). Thus, use of an independent shear-wave
measurement obtained from SV-P data in conjunction with
P-P gathers helps constrain the inversion result of IS, making
IS estimate more reliable compare to IS obtained from P-P
data alone.
Conclusions
This study investigates the ability of SV-P data generated by
available P-wave sources and recorded by vertical
geophones to substitute for P-SV data recorded by 3C
geophones and to increase the value of conventional P-wave
data for subsurface characterization. We demonstrate
through comparison of P-P and SV-P seismic data and
inverted elastic properties that vertical-vibrator SV-P data
from vertical geophones works as good as P-SV seismic data
from horizontal geophones to understand reservoir
properties, and provide a better resolution compared to SV-
P data generated by horizontal vibrator. This study shows
that the use of SV-P data from a vertical vibrator, along with
P-P data, significantly improves S-impedance estimates in
the Arbuckle injection zone. Such improved S-impedance
Figure 6: (a) P-impedance (IP) obtained from joint-inversion of P-
characterization helps to delineate weak and fractured
P gathers and SV-P stacks (from vertical vibrator) for Line-1
compared to IP computed from well-logs (vertical colored strips injection zones in the Lower Arbuckle from tight and
below well labels). High IP in Middle Arbuckle (between A and B), siliceous baffle-zones present in the Middle Arbuckle.
and low IP in Lower Arbuckle injection zone are consistent with the Independent shear-wave measurements made with the SV-P
observations made on IP obtained from P-P AVO inversion. (b) S- mode utilizes low-cost, conventional P-wave seismic data
impedance (IS) obtained from joint-inversion of P-P gathers and acquisition setups, Converted-wave studies can now be
SV-P stacks (from vertical vibrator) for Line-1 compared to IS performed on available legacy P-wave seismic data, which
computed from well-logs. IS estimated from P-P and SV-P joint will revolutionize the way that multi-component reflection
inversion better resolves lithology variation in Kansas City and
seismology is done.
Mississippi horizons, and within Arbuckle Group, compared to IS
estimated from P-P inversion (Figure 5b). Lower Arbuckle injection
zone (below B) shows low values of IS, suggesting lithology Acknowledgements
variation not as evident on P-P inversion results. We thank Dr. W. Lynn Watney and Dr. Jason Rush from the
Kansas Geological Survey for providing the Wellington
Next, P-P and SV-P data are jointly inverted to see if using Field dataset and Bruce Karr and Adrian Chen from
SV-P data adds value to subsurface property estimation. FairfieldNodal for seismic data processing. Support for this
Figure 6 shows the results from joint inversion of P-P gathers work was provided by the Bureau of Economic Geology at
and SV-P stacks (from vertical vibrator) for Line-1. Figure The University of Texas at Austin. Hampson-Russell
6a is the inverted IP with color-fill inserted at the well Software and Services provided software for basic multi-
locations being the IP computed from well-logs. High IP in component seismic interpretation via their University Grant
the Middle Arbuckle (between horizons A and B), and low Program.
IP in the Lower Arbuckle injection zone are consistent with
the observations made on IP obtained from P-P AVO
© 2017 SEG Page 5185

EDITED REFERENCES
REFERENCES
Byrnes, A. P., E. K. Franseen, and D. M. Steinhauff, 1999, Integrating plug to well-scale petrophysics
with detailed sedimentology to quantify fracture, vug, and matrix properties in carbonate
reservoirs — An example from the Arbuckle Group, Kansas: Kansas Geological Survey, Open-
File Report 99-47.
Franseen, E. K., A. P. Byrnes, J. R. Cansler, D. M. Steinhauff, T. R. Carr, and M. K. Dubois, 2003,
Geologic controls on variable character of Arbuckle reservoirs in Kansas — An emerging picture:
Kansas Geological Survey, Open-File Report 59, 30.
Gaiser, J., and R. Verm, 2012, SS-wave reflections from P-wave sources in azimuthally anisotropic
media: 82nd Annual International Meeting, SEG, Expanded Abstracts, 1–5,
Gupta, M., M. V. DeAngelo, and B. Hardage, 2015, P-P and S-S Wave interpretation of a carbonate
formation: A case study from the Arbuckle interval in Wellington Field, Kansas: 85th Annual
International Meeting, SEG, Expanded Abstracts, 2082–2087,
Gupta, M., K. Spikes, M. Far, D. Sava, and B. Hardage, 2014, Statistical AVO intercept-gradient analysis
of direct S-waves: A methodology for quantitative fracture characterization: 84th Annual
International Meeting, SEG, Expanded Abstracts, 506–511, https://doi.org/10.1190/segam2014-
0542.1.
Gupta, M., K. Spikes, and B. Hardage, 2016, Characterization of naturally fractured Arbuckle Group in
the Wellington Field, Kansas, using S-wave amplitude variation with offset: Interpretation, 5,
T49–T63, https://doi.org/10.1190/int-2016-0061.1.
Hardage, B. A., and D. E. Wagner, 2014a, S-S imaging with vertical-force sources: Interpretation, 2,
SE29–SE38, https://doi.org/10.1190/int-2013-0097.1.
Hardage, B. A., and D. E. Wagner, 2014b, Generating direct-S modes with simple, low-cost, widely
available seismic sources: Interpretation, 2, SE1–SE15, https://doi.org/10.1190/int-2013-0095.1.
Interpretation, 3, SW27–SW35, https://doi.org/10.1190/int-2015-0002.1.
Ohl, D., and A. Raef, 2014, Rock formation characterization for carbon dioxide geosequestration: 3D
seismic amplitude and coherency anomalies, and seismic petrophysical facies classification,
Wellington and Anson-Bates Fields, Kansas, USA: Journal of Applied Geophysics, 103, 221–
231, https://doi.org/10.1016/j.jappgeo.2014.01.017.
Watney, W. L., and J. Rush, 2012, Small scale field test demonstrating CO2 sequestration in Arbuckle
saline aquifer and by CO2-EOR at Wellington field, Sumner County, Kansas: DOE Project
Report Number DE-FE0006821, 28.
© 2017 SEG Page 5186


Examples of SV-P Images Made with P Sources and Vertical Geophones: Bob A. Hardage*,
Bureau of Economic Geology, Jackson School of Geosciences, Austin, TX
Introduction
Scientists at the Bureau of Economic Geology (BEG) have worked with a small group of seismic data-processing companies to
create SV-P images of selected reservoir targets from legacy P-wave data. In this discussion, legacy P-wave data are data that
have been generated by a P source, recorded with vertical geophones, and preserved in seismic data libraries for a number of
years before SV-P data processing is initiated. These data can be either surface-recorded data or VSP data. In the case of VSP
data, the SV-P mode is embedded in the same rotated-geophone response that is used to create a P-P image. There is widespread
interest among people who own legacy P-wave data to determine if valuable S-mode images and S-wave attributes can be
generated for reservoir characterization via the SV-P mode that, although embedded in the vertical-geophone data they own, has
been ignored as a low-cost (almost free) option for practicing S-wave reflection seismology. This paper will summarize the work
that BEG has done to determine the value of SV-P data produced by common P sources.
Ellenburger karst: Midland Basin
The Ellenburger Formation is a thick, regional, carbonate system that extends across a large region of Texas. A unique feature of
the Ellenburger is that it contains a variety of karst features in many geographic locations. An opportunity arose whereby SV-P
imaging of the Ellenburger could be investigated in an area on the western shelf of the Midland Basin in west Texas. The data
that were used were recorded 12 years previous to this SV-P imaging effort. The data were recorded by 3C geophones, which
allowed P-SV images to be created from horizontal-geophone data and compared with SV-P images extracted from vertical-
geophone data. At this study location, the depth of the Ellenburger was approximately 16,000-ft (4900-m). Side-by-side displays
of P-P, SV-P, and P-SV images that span the Ellenburger which were created in this investigation are displayed in Figure 1. An
important finding was that the SV-P image at this location shows karst evidence that cannot be readily seen in either the P-P or
the P-SV images.
Strawn conventional reservoirs: Midland Basin
The Strawn is a carbonate-dominated interval positioned above the Ellenburger Formation discussed above (see formation tops in
Figure 1). Productive Strawn reservoirs having porosities of 10 to 15 percent are sometimes encountered across the Midland
Basin. A characteristic of many of these conventional Strawn reservoirs is that they are long-lived producers that can be operated
for several decades. Operators are always searching for such reservoirs. An opportunity to image geology associated with three
particularly attractive Strawn producers with SV-P data arose, and this opportunity again allowed a vertical-geophone SV-P
image to be compared with its horizontal-geophone equivalent, a P-SV image. This image comparison is shown in Figure 2. In
discussing these images, the author will show proof that the P-SV image contains a strong acquisition footprint, but no evidence
of an acquisition footprint can be found in the SV-P image. Anomalous reflection amplitudes associated with this acquisition
footprint decrease the value of P-SV data for identifying conventional Strawn reservoirs in this particular seismic survey.
Wolfberry unconventional turbidite reservoirs: Midland Basin
The Wolfberry interval of the Midland Basin, which begins with the Spraberry Formation and extends down to the base of the
Wolfcamp, is one of the most prolific, oil-producing, unconventional reservoir systems in the U.S. In the Midland Basin, the
Wolfberry interval has a thickness of approximately 5000-ft (1500-m) in the central portion of the basin and thins to
approximately 2000-ft (600-m) on the basin shelf. As many as 10 oil-bearing, low-permeability, unconventional-reservoir
systems can be encountered within the Wolfberry interval. The Wolfberry was deposited as a massive stack of fine-grain
turbidites in which mineral mixtures of turbidite flows vary laterally and vertically in unpredictable patterns across the basin shelf
and into the deep basin (Hamlin and Baumgardner, 2012). The author will show stratal slices of 3D SV-P semblance volumes of
the Wolfberry system that illustrate, in map views, depositional features that are identical to features expected of turbidite
deposition; i.e., broad, elongated, irregular-shaped geobodies that have lateral dimensions ranging from 0.5 to 3 km, but a
maximum vertical thickness of only 10 to 30-m. These SV-P turbidite details will be described in the oral presentation.
Marcellus Shale: Appalachian Basin
A comprehensive seismic database in the Appalachian Basin that imaged the Marcellus Shale and its associated geology became
available and allowed a valuable principle of SV-P imaging to be illustrated. These data consisted of a 3C3D surface survey and
an elaborate VSP program that provided walk-around, walk-away, and variable-offset VSP data within the 3D seismic image
space. The VSP data provided an opportunity to extract a SV-P image from the same rotated-geophone data that were used to
make a conventional VSP P-P image. This VSP-based SV-P image was then compared with the P-SV image constructed from the
surface-recorded 3C3D data that surrounded the VSP well. This comparison established two important principles: (1) SV-P and
P-SV images are identical if data-acquisition and data-processing are properly done, and (2) SV-P images can be extracted from
1

© 2017 SEG Page 5187

VSP data as well as from surface-recorded data (Li and Hardage, 2015). An image comparison of these two genetically-related
converted modes (SV-P and P-SV) is displayed in Figure 3.
Midcontinent (Kansas) CO2 storage reservoir

The Kansas Geological Survey provided 2D9C data, core analyses, and well-log data from a calibration well across a CO2
storage site in Sumner County, Kansas, to support a PhD student thesis study supervised by the author. These Kansas data
provided an opportunity to compare a SV-P image created by a vertical-vibrator with the same SV-P image created by a
horizontal-vibrator. To the author’s knowledge, no such image comparison existed in the literature until Menal Gupta’s PhD
thesis was published at the University of Texas at Austin (Gupta, 2017). This seminal comparison of SV-P images generated by
vertical and horizontal vibrators is displayed in Figure 4. In general, the images are identical, but some geological units are
portrayed with different amplitude strengths and different frequency spectra in the two images. The SV-P image produced from
vertical vibrator data has a broader spectral frequency than the SV-P image made by the horizontal vibrator. Work is ongoing to
determine which SV-P reflection amplitude behavior, that produced by a vertical vibrator or that produced by a horizontal
vibrator, is the better correlation with rock and fluid attributes at this CO2 storage reservoir.
Conclusions
The SV-P images shown in this discussion will illustrate that it is possible to create valuable S-mode images and to extract
valuable S-wave attributes from data generated by a P source and recorded with vertical geophones. The cost of acquiring SV-P
data in this manner is zero because SV-P data are automatically acquired whenever P-P data are acquired. It appears the
excessive cost of S-wave data acquisition that has held back the use of S-wave reflection seismology is now no longer a barrier.
There are, no doubt, some earth-surface conditions where it will be difficult to create acceptable quality SV-P data with P
sources. Only expanded use of the concept of utilizing the direct-SV wavefields produced by P sources will help seismic data
users understand what earth-surface conditions and/or what deep geology conditions are not favorable for practicing SV-P
reflection imaging with P sources.
Figure 1. (a) Comparison of depth-equivalent P-P, SV-P, and P-SV images of the Ellenburger interval in west Texas. (b)
Evidence of karst-collapse shown by solid lines on the SV-P profile is difficult to find in the companion P-P and P-SV profiles.
2

© 2017 SEG Page 5188

Figure 2. SV-P and P-SV images of conventional Strawn reservoirs. The two images should be identical but are obviously
different. Investigation showed that the P-SV image is contaminated with an acquisition footprint, but the SV-P image is not.
Figure 3. (a) Comparison of VSP SV-P image with a P-SV image constructed from surface-based 3C3D data surrounding the
VSP well. (b) Isolated view of the SV-P image. VSP data provided by Chesapeake Energy. Multiclient surface data presented
with permission from Geophysical Pursuit Inc and Geokinetics Inc. Modified from Li and Hardage (2015)
3

© 2017 SEG Page 5189

Figure 4. Comparison of SV-P images made by (a) a vertical vibrator, and by (b) a horizontal vibrator across a CO2 storage
reservoir. Modified from Gupta, 2017. Data provided by the Kansas Geological Survey.
4

© 2017 SEG Page 5190
EDITED REFERENCES
REFERENCES
Gupta, M., 2017, Using direct S-wave seismic modes for reservoir characterization in Wellington, Field,
Kansas: PhD thesis, The University of Texas at Austin.
Hamlin, H. S., and R. W. Baumgardner, 2012, Wolfberry (Wolfcampian-Leonardian) deep-water
depositional systems in the Midland Basin–stratigraphy, lithofacies, reservoirs, and source rocks:
Report of Investigations No. 277, Bureau of Economic Geology, Austin.
Interpretation, 3, SW27–SW35, http://dx.doi.org/10.1190/INT-2015-0002.1.
© 2017 SEG Page 5191

Surface-wave imaging of inversely dispersive media: a permafrost example
S. Dou*1, J.B. Ajo-Franklin1, B. Dafflon1, J. Peterson1, C. Ulrich1, D. Dreger2, S.S. Hubbard1
1
Lawrence Berkeley National Laboratory, 2University of California, Berkeley
Summary minimize the mismatch between the observed and the

model-predicted dispersion spectra. The forward solver
Surface waves are effective in identifying inversely computes the complete wavefield with wavenumber
dispersive media, but because dispersion-curve retrieval is integration method (Herrmann 2004). The nonlinear inverse
susceptible to ambiguities, commonly-used inversion problem is solved using particle swarm optimization (PSO)
methods often become inapplicable. Here we highlight a (Eberhart and Kennedy 1995), a derivative-free global
full-wavefield approach that uses dispersion spectra instead search method that is readily parallelizable and fast-
of dispersion curves to invert for shear-wave velocity (VS) converging. We demonstrate the effectiveness of the method
structures. With a case study of permafrost imaging, we with a case study of permafrost imaging, in which
demonstrate that this method is able to achieve optimal data pronounced low-velocity layers are identified and mapped
fits without requiring good starting models. This makes the with surface waves.
spectrum-fitting inversion superior to waveform-fitting
approaches when little prior knowledge is available. This
study also takes advantage of particle swarm optimization
(PSO), a derivative-free global search method that is suitable
for but not yet widely used in solving nonlinear inverse
problems in geophysics.
Introduction
Inversely dispersive media is characterized by the presence

of embedded low-velocity layers/zones in a high-velocity
background. Such velocity structures cannot be effectively
mapped with seismic refraction, as refracted body waves are
known to be “blind” to velocity reversals. By contrast,
surface waves can exhibit inverse dispersion patterns that are
strikingly different from their normally dispersive
counterparts. Thus, surface-wave methods are uniquely
suitable for seismic imaging of inversely dispersive media.
The utility of surface waves for identifying inversely

dispersive media is apparent, but the subsequent inversion
for shear-wave velocity (VS) structures is not trivial. This is
because most established inversion methods use dispersion
curves as the data, whereas reliable retrieval of dispersion
curves from inversely dispersive waves is rarely feasible due
to strong excitations of higher-order modes. Although strong
Figure 1: Synthetic example illustrating strong higher modes in
higher modes can also be observed in normally dispersive inversely dispersive surface waves. (a) Synthetic shot gather in
wavefields, they pose a bigger challenge in the case of offset-time (x-t) domain. (b) Inversely dispersive input model. (c)
inverse dispersion because higher modes can be so closely Dispersion spectrum overlain by modal curves (color coded by
spaced in the dispersion spectrum that several modes often mode numbers). VS = shear-wave velocity.
emerge as one high-amplitude “patch” (Figure 1). This
plagues dispersion-curve retrieval with ambiguities and Background, field site, and acquisition parameters
consequently renders conventional inversion methods
inapplicable. VS structures of permafrost can be highly irregular, as the
influences of water/ice content variations are often stronger
In this study, we highlight a full-wavefield approach for than those of the lithology changes alone. Velocity reversals
surface-wave inversion of inversely dispersive media. are particularly common in saline permafrost that is
Instead of using dispersion curves, this method aims to widespread in coastal areas of the Arctic. The combined
© 2017 SEG Page 5192

Surface-wave imaging of inverse dispersive media
effects of subzero air temperatures, freezing-point points in the dispersion spectrum. Note that the computation
depression of dissolved salts, and salt exclusions of ice can of the model-predicted dispersion spectrum S requires
yield a subsurface configuration in which partially- forward modeling of the full wavefield. This inevitably
frozen/unfrozen, low VS layers are overlain by frozen, high increases computational costs of the inversion.
VS layers.
Our field site in Barrow, Alaska contains saline permafrost

(Figure 2a). We first observed inversely dispersive surface
waves along a 470-meter-long transect in the winter of 2012
(Dou and Ajo-Franklin 2014). Two years later in 2014, we
extended the survey transect to 4300 meters so as to
determine if saline permafrost is pervasive across the Barrow
Environmental Observatory (BEO) (Figure 2b). We
acquired four nearly equidistantly distributed shot gathers
along the transect with a sledge hammer source (30 lbs) and
a geophone array of 1 meter spacing (10 Hz vertical
geophones). Each shot gather covers an offset range of 84
meters.
Figure 2: Map view of the permafrost field site near Barrow,

Alaska. (a) Location of Barrow, Alaska and (b) the 4300-meter-long
suvey transect (red line) across the Barrow Environmental
Observatory (BEO). Red stars (1–4) denoites the locations of the
multichannel surface-wave surveys.
Figure 3: Inversely dispersive field data. Left: shot gathers in offset-
time (x-t) domain. Right: dispersion spectra in frequency-velocity
Inversely dispersive field data (f-v) domain. (a), (b), (c), and (d) correspond to line 1, 2, 3, and 4 in
Figure 2b.
Figure 3 shows field data acquired along the 4300-meter-
long BEO transect. The dispersion spectra, though differ in Forward modeling and model parameterization
detail, all show inversely dispersive trends in which high-
frequency spectral highlights correspond to higher phase Forward modeling of the synthetic shot gather requires
velocities than those of the low-frequency highlights. convolving a source wavelet with the impulse response of
the model (i.e., the Green’s function). We use zero-offset
Spectra-based objective function traces of the field data as empirical source wavelets. The
impulse response of a layered model is computed with the
The objective function measures the root-mean-square- wavenumber integration method of Herrmann (2004).
deviation (RMSD) between the model-predicted (S) and the
observed (O) dispersion spectra as follows: Because computational costs of the forward modeling are
high, parsimonious parameterization is adopted to ensure
-1 -. timely convergence of the inversion. We set up a three-layer
'/0 (/0(𝑆'( − 𝑂'( ),
𝑀 𝒎 = configuration in which the top layer and the half-space
𝑁3 𝑁4 basement have uniform velocities. The mid layer is
parameterized using a cubic Hermite spline with four nodes
where m is the vector of model parameters; M is the RMSD (Figure 4). We use first-arrival velocities as VP of the
misfit; Nf and Nv are the number of frequency and velocity topmost layer. We also fix VP/VS and density ρ to further
© 2017 SEG Page 5193

reduce the amount of unknowns. In the end, we are left with The bird flock forms a swarm, within which each bird is a
nine unknowns as shown in Table 1. particle. The location of the food is analogous to the global
optima of the inverse problem. At any kth iteration of the

search process, each particle explores the search space with
(B)
its position 𝑝' (𝑖 = 1, 2, ⋯ , 𝑁) influenced both by its own
best past position 𝑝'H;IJ and the best past position of the
swarm 𝑝KH;IJ . The particle velocity used for updating the
particle position is computed via
(B) BN0
𝑣' = 𝑤 (B) 𝑣 (BN0) + 𝑐Q 𝑟0 𝑝'H;IJ − 𝑝' + 𝑐K 𝑟, (𝑝KH;IJ
(BN0)
− 𝑝' )
where 𝑤 (B) is the inertia coefficient that controls the

exploration ability of the swarm, with larger 𝑤 (B) promoting
wider exploration of the search space, and vice versa; 𝑐Q is
cognitive learning factor that controls the biases toward the
personal best 𝑝'H;IJ ; 𝑐K is the social learning factor that
controls the biases toward the best position of the swarm
𝑝KH;IJ ; 𝑟0 and 𝑟, both are random numbers between 0 and 1.
Once the particle velocity is obtained, each particle’s
(B) (BN0) (B)
position is updated with 𝑝' = 𝑝' + 𝑣' .
In our inversion, we distribute 23 particle search in parallel

Figure 4: Model parameterization illustration. Yellow dots are the on 23 processors (1.2 GHz Genuine Intel processor). Each
nodes of the cubic Hermite spline in the mid layer. PSO run went through 138–392 search steps over the wall-
clock time of 3–8 hours (1 to 1.5 minutes per search step).
Table 1: Model parameters. For a 9-variable inversion, such computation cost is high
Layer Characteristics Variables despite the use of the parallelized PSO. The high costs are
index mainly because of forward modeling of the full wavefield.
1 Uniform layer (1) ℎ0
2 Cubic Hermite Node 1: (2) 𝑉7, 89:;0 Inversion results
89:;,
spline with Node 2: (3) 𝑉7, (4) ℎ,89:;,
four nodes 89:;<
Node 3: (5) 𝑉7, (6) ℎ,89:;<
89:;=
Node 4: (7) 𝑉7, (8) ℎ,89:;=
3 Uniform half- (9) 𝑉7<
space
- h is layer thickness (in m); VS is S-wave velocity (in
m/s).
- Density of all layers is fixed at 2000 kg/m3.
- P-wave velocities (VP) of the top layer are the velocities
?'8;0 ?'8;,
of the first arrivals: 𝑉>0 =3186 m/s; 𝑉>0 =3079 m/s;
?'8;< ?'8;=
𝑉>0 =3410 m/s; 𝑉>0 =3307 m/s.
- 𝑉> 𝑉7 = 1.732 when 𝑉7 ≥1500 m/s; 𝑉> 𝑉7 = 4.0
otherwise. Figure 5: Ensembles of top 10% best-fitting VS profiles obtained
from PSO. Black, bold profile denote the global optima. Rest of
Particle swarm optimization (PSO) profiles are color-coded with misfit reduction computed relative to
the global optima. Numerical labels 1, 2, 3, and 4 correspond to line
1, 2, 3, and 4 in Figure 2b.
We use a MATLAB implementation of particle swarm
optimization (PSO) algorithm (Chen 2014) to solve the
We compile the inversion results in Figure 5. Optimal VS
nonlinear inverse problem. PSO is a population-based
profiles obtained from the four survey lines all show the
stochastic global optimization method. Its working principle
same first-order feature: A low-velocity mid layer is
can be compared to a flock of N birds searching for food.
© 2017 SEG Page 5194

Figure 6: Data fits in space-time (x-t) domain and frequency-velocity (f-v) domain. In x-t domain, synthetic and observed shot gathers are
displayed in red and gray, respectively. (a), (b), (c), and (d) correspond to line 1, 2, 3, and 4 in Figure 2b.
sandwiched between a high-velocity, thin top layer (1–5.7

meter thick) and an intermediate-velocity half-space. VS of Surface waves are uniquely suitable for imaging inversely
the mid layer, which fall within the range of 301–999 m/s, dispersive media, but the difficulties in reliably retrieving
are 44%–84% lower than VS of the topmost frozen layer. dispersion curves also render most inversion methods
Figure 6 shows the data fits provided by the PSO global inapplicable. By using a full-wavefield approach that uses
optima. Note that although the inversion is conducted dispersion spectra instead of dispersion curves in the
entirely in the frequency-velocity (f-v) domain, the best-fit inversion, we are able to obtain VS profiles that provide good
models are able to achieve good data fits in the space-time data fits both in space-time (x-t) domain and frequency-
(x-t) domain. velocity (f-v) domain. This study also demonstrates the
effectiveness of PSO algorithm in solving nonlinear inverse
Conclusions problems.
© 2017 SEG Page 5195

EDITED REFERENCES
REFERENCES
Chen, S., 2014, Another particle swarm toolbox (in MATLAB),
http://www.mathworks.com/matlabcentral/fileexchange/25986-another-particle-swarm-toolbox,
accessed 12 Nov., 2014.
Dou, S., and J. B. Ajo-Franklin, 2014, Full-wavefield inversion of surface waves for mapping
embedded low-velocity zones in permafrost: Geophysics, 79, EN107–EN124,
http://dx.doi.org/10.1190/geo2013-0427.1.
Eberhart, R. C., and J. Kennedy, 1995, A new optimizer using particle swarm theory: Proceedings of the
Sixth IEEE International Symposium on Micro Machine and Human Science,
http://dx.doi.org/10.1109/MHS.1995.494215.
Herrmann, R. B., 2004, Computer programs in seismology, http://www.eas.slu.edu/eqc/eqccps.html,
accessed 15 Jan., 2012.
© 2017 SEG Page 5196

P- and S-wave velocity models from surface wave dispersion curves data transform
Laura Valentina Socco*, Farbod Khosro Anjom, Politecnico di Torino, Italy; Cesare Comina, Daniela Teodor,
Università degli Studi di Torino, Italy.
Summary models with good approximation. Here we will extend the

interval velocity estimation to VP and show the application
The inversion of surface wave dispersion curves is a well- to a field dataset.
established investigation approach to provide S-wave
velocity models. Recent works have shown the possibility Method
of exploiting the relation between surface wave wavelength
and investigation depth to estimate directly the time- The method (Socco et al., 2017) is explained in the
average S-wave velocity along a line or over an area. following using a synthetic 1D velocity model (Figure 1). It
Moreover, the same relation is sensitive to Poisson’s ratio is based on the estimation of a relationship between SW
and can be therefore used to retrieve also the time-average wavelength and investigation depth. In particular, starting
P-wave velocity model. Here we show that, starting from from a known VS model and its relevant dispersion curve,
the estimated time average VS and VP models and using a first, the VS model is transformed into VSz:
Dix-like formula, we can estimate local VS and VP models.
We use a synthetic 1D example to outline the method and ∑ hi (1)
then we show the results on a field dataset. VSz = n
h
∑ VSi
Introduction n i
The inversion of surface wave (SW) dispersion is routinely where n is the number of layers down to depth z, and hi and
used to retrieve local 1D S-wave velocity models for VSi are the thickness and the velocity of the ith layer,
various exploration and engineering applications (Socco et respectively. Then, for each value of z, we search for the
al., 2010). S-wave velocity (VS) is a key parameter for wavelength at which the phase velocity of surface waves
several engineering problems ranging from geotechnical (green in Figure 1) is equal to the VSz at that depth (blue
characterization to seismic hazard. Nevertheless, for several diamonds in Figure 1). The wavelength-depth (W/D)
purposes, among which static corrections for seismic couples (red in Figure 2) are used to estimate the time
reflection processing and aquifer characterization, P-wave average VSz (blue dots in Figure 1).
velocity model is also required. It is hence desirable to
implement novel methods to retrieve also P-wave velocity
(VP) from surface wave data.
In the usual SW analysis, due to poor sensitivity of
dispersion curve to VP, only VS is considered unknown in
the inversion process and VP (or Poisson’s ratio, ν) is
assumed on the basis of a priori information, hence, SW
analysis is not considered suitable for a reliable estimation
of both VS and VP.
Recent works (Socco et al., 2017) have introduced a novel
approach to estimate time-average VS (VSz) over a seismic
line using a relationship between SW wavelength and
investigation depth which is estimated for a single
reference profile along the line. This method proved to be
reliable for estimating VSz also in case of lateral variations.
Moreover, the relationship used for the estimation of VSz
proved to be sensitive to Poisson’s ratio (Socco and
Comina, 2017) and was hence successfully used to estimate
Poisson’s ratio and transform the VSz profiles into VPz Figure 1 – VS and VP model (blue and red solid), the
corresponding SW dispersion curve plotted as a function of
profiles. wavelength (green), the corresponding VSz and VPz (blue and red
Socco and Comina (2015) showed on synthetic data that, diamonds), the VSz and VPz estimated using the W/D in Figure 2
starting from VSz profiles along a line it was possible, using (blue and red dots), estimated VS and VP from estimated VSz and
VPz (blue and red dashed lines).
a Dix-like equation, to obtain local (interval) velocity
© 2017 SEG Page 5197

P- and S-wave velocity models from surface wave data transform
The W/D depends on Poisson’s ratio ν (Socco and Comina, Field data example
2017) and hence it can be used to estimate ν at each depth

by comparison with synthetic W/D for models with The field data were acquired in 2007 on an alpine fan in the
constant ν (blue lines in Figure 2). The estimated ν allows north-west of Italy (see Socco et al., 2008 for details about
the estimated VSz to be transformed into VPz (red dots in acquisition and processing). The dispersion curves were
Figure 1). extracted from the records acquired for high resolution
seismic reflection line about 1 km long. These dispersion
curves were inverted using a laterally constrained inversion
(LCI) code (Socco et al., 2009) and provided a pseudo-2D
VS model. The same records were also used to pick P-wave
travel times and carry out a P-wave tomographic inversion.
The VSz and VPz models were estimated, with the
procedure outlined above, in Socco et al., (2017) and Socco
and Comina (2017) using a reference model coming from a
passive + active SW survey carried out close to the seismic
line. Here we use equation 2 to transform the VSz and VPz
models into VS and VP and we compare them with the
results of LCI and P-wave tomography as benchmarks. We
also compare our results with the velocity models obtained
from a down hole test (DH) performed in a 50 m deep well
close to the seismic line. In Figure 3 we show the DH VS
and VP models (solid blue and red line respectively) and
we compare them with our estimate (blue and red dots) and
benchmark profiles extracted from LCI and P-wave
tomography in the same position (blue and red dashed line).
Figure 2 – The wavelength/depth relationship for the model in

Figure 1 (red dots) compared with synthetic W/D for the same VS
model and different values of Poisson’s ratio (blue line).
Using the estimated VSz and VPz we can then apply a Dix-
like equation and find the local velocity values:
zn − zn−1 (2)
Vn =
zn z
− n−1
Vzn Vzn−1
where Vn is the velocity (either VS or VP) of the nth

considered layer, zn and zn-1 are the depth of the bottom
and the top of the nth layer, respectively, and Vz,n and
Vz,n-1 are the time-average velocities (either VSz or VPz) at
the bottom and the top of the nth layer. The estimated local
velocities, computed using a layer thickness of 0.5m, are
shown in Figure 1 (dashed blue and red lines) and provide a
smooth reconstruction of the true velocity model. Figure 3 – Comparison among DH results, VS and VP models
Socco et al. (2017) and Socco and Comina (2017) showed obtained through our method and those from the benchmark (SW
that the W/D relationship, and corresponding Poisson’s LCI for VS and P-wave tomography for VP) in position close to the
DH test.
ratio, retrieved for a single VS model within a dataset can
be used to transform all the dispersion curves in the dataset
The comparison with DH test shows that both VS and VP
in VSz models and then in VPz. In the following we show a
are retrieved with good approximation. Nevertheless the
field data example and we compare the final VS and VP
estimate of VP in deeper part of the model shows high error
models with benchmarks.
with respect to DH and also significant error with respect to
P-wave tomography benchmark.
© 2017 SEG Page 5198

Figure 4 – VS models: a) benchmark from LCI, b) estimated from estimated VSz, c) normalized error between benchmark and estimated VS. The
arrows indicate the reference model (RM) and the DH test locations along the line; the white dots indicate the experimental data points expressed
as a function of wavelength.
equation 2. They indicate that further analyses are worth to
It has to be considered that the position of the reference improve the reliability of the method.
model used to estimate the W/D and the Poisson’s ratio is
on the opposite side of the line with respect to DH tests (see Conclusions
arrows on Figures 4 and 5).
The comparison of the results along the whole seismic line We have shown a novel method to estimate VS and VP
is shown in Figure 4 and Figure 5 for VS and VP, models through a data transform of a set of experimental
respectively. The normalized errors are also shown. The SW dispersion curves. The only information introduced in
results show that the errors are mostly within 15 % with the process is a single VS model in correspondence of a SW
some zones with very large errors on the deeper part of the dispersion curve. This reference model can be obtained by
models. These errors occur in zones of lower data coverage inverting one of the dispersion curves in the dataset or from
and are related to the high sensitivity to noise of the other a priori information.
© 2017 SEG Page 5199

Figure 5 – VP models: a) benchmark from P-wave tomography, b) estimated from estimated VPz, c) normalized error between benchmark and
estimated VP; the arrows indicate the reference model (RM) and the DH test locations along the line; the white dots indicate the experimental
data points expressed as a function of wavelength.
knowledge. The actual method, based on the estimation of
The model, together with the relevant dispersion curve, is Poisson’s ratio through comparison of the experimental
used to estimate the wavelength/depth relationship (W/D) W/D relation and theoretical W/D for different values of
that represents the function for transforming the dispersion Poisson’s ratio, is very simple and, when it uses the
curves directly into time-average VS models (VSz). The Poisson’s ratio estimated at the reference model to retrieve
same relationship is used to estimate Poisson’s ratio that VP along the whole line, it makes the assumption of
represents the function for transforming the VSz into VPz. negligible lateral variations of Poisson’s ratio. To avoid
By applying a Dix-like equation the time-average velocity such assumption and to improve the accuracy of the results,
models are transformed into VS and VP models along the inversion using W/D relationship to estimate VP is the most
line. This approach represents, to our knowledge, the first obvious evolution of the method. The advantage of the
successful method for estimating both VS and VP from actual data transform approach lies in its simplicity and
surface wave dispersion curve and without any a priori absence of any arbitrary parameters to be set.
© 2017 SEG Page 5200

EDITED REFERENCES
REFERENCES
Socco, L.V., D. Boiero, C. Comina, S. Foti, R. Wisén, 2008, Seismic Characterization of an Alpine Site:
Near Surface Geophysics, 6, 255–267, http://dx.doi.org/10.3997/1873-0604.2008020.
Socco, L.V., D. Boiero, S. Foti, R. Wisén, 2009, Laterally constrained inversion of ground roll from
seismic reflection records: Geophysics, 74, G35–G45, http://dx.doi.org/10.1190/1.3223636.
Socco, L.V., C. Comina, 2015, Approximate direct estimate of S-wave velocity model from surface wave
dispersion curves : 77th Annual International Conference and Exhibition, EAGE, Extended
Abstracts, Tu 21 A09, http://dx.doi.org/10.3997/2214-4609.201413714.
Socco L.V., C. Comina, F. Khosro Anjom, 2017, Time-average velocity estimation through surface-wave
analysis: Part 1 — S-wave velocity: Geophysics, 82, U49–U59,
http://dx.doi.org/10.1190/GEO2016-0367.1.
Socco L. V. and C. Comina, 2017, Time-average velocity estimation through surface-wave analysis: Part
2 — P-wave velocity: Geophysics, 82, U61-U73, http://dx.doi.org/10.1190/GEO2016-0368.1.
© 2017 SEG Page 5201

Application of active and passive surface wave methods to shallow to deep S-wave velocity
estimation
Koichi Hayashi*, OYO Corporation
Summary dimensional (1D) VS profiles down to a depth of 100 m

(Nazarian et al., 1983). SASW surveys rely on a controlled
The use of surface wave methods in S-wave velocity (VS) source such as a shaker or vibrator, from which
estimation has been getting popular during the past few practitioners calculate phase differences between the two
decades. This article briefly reviews various seismic receivers via cross-correlation.
methods using active and passive surface waves and
discusses their applicability to site investigations. This Park et al. (1999a, 1999b) proposed a multichannel analysis
discussion particularly focuses on the applicability of the of surface waves (MASW) method, whereby practitioners
passive surface wave methods in comparison with active transform the multichannel surface wave waveform data
methods. As passive methods are relatively new, the pros from the time-distance domain into the phase velocity-
and cons of the technique are not yet public knowledge. We frequency domain, which they then use to determine phase
discuss penetration depth of active and passive surface velocities. MASW is superior to SASW for recognition of
wave methods from the maximum wave length point of dispersion curves and distinguishing the fundamental mode
view. Empirical data show that the penetration depth of the Rayleigh wave from other modes, such as higher modes
passive surface wave methods using SPAC is and body waves. Xia et al. (1999) and Miller et al. (1999)
approximately equal to array size or maximum separation applied MASW to shot gathers along a survey line and
of receivers. Applicability of irregular arrays in the passive delineated pseudo two-dimensional (2D) VS sections.
surface wave methods is also discussed with field examples. Hayashi and Suzuki (2004) applied common depth-point
The results imply that ambient noise is generally (CDP) analysis as used in 2D seismic reflection surveying
omnidirectional in the investigation area and irregular to MASW, and utilized common midpoint cross-correlation
arrays can provide reliable information at most sites. (CMPCC) analysis to increase the lateral resolution of the
surface wave methods in heterogeneous environments.
Geophysical methods using surface waves
During the past few decades, researchers have made
During the past few decades, there has been significant considerable progress towards the development of passive
development in the use of surface wave methods for S- surface wave methods utilizing ambient noise or
wave velocity (VS) estimation. Table 1 summarizes the microtremors. The methods are typically called
active and passive surface wave methods that are currently microtremor array measurements (MAM) since two
the most widely used. A spectral analysis of surface waves dimensional arrays are usually used for calculating phase
(SASW) has been used for the determination of one- velocity from ambient noise. Aki (1957) investigated
Table 1: Various active and passive surface wave methods.
© 2017 SEG Page 5202

Active and passive surface wave methods
ambient noise as surface waves and proposed a theory of eliminate heavy cables. The benefits of the methods,
spatial autocorrelation (SPAC). Okada (2003) developed a however, are not currently public knowledge since the
large scale passive surface wave method or microtremor passive methods are relatively new compared to the active
array measurements (MAM) based on SPAC in order to methods. In this discussion, we are going to focus on two
estimate deep VS structure. different observations used in the SPAC method.
The FK method is another popular method for calculating Array size and investigation depth in SPAC
the phase velocity from the ambient noise. For many years,
researchers have widely used both SPAC and FK. There is Here we discuss the relationship between array size and
fairly general agreement that SPAC works with fewer penetration depth. Figure 1 compares minimum and
receivers and more irregular arrays compared with FK. As maximum wavelength obtained from active and passive
a result, in recent years SPAC is becoming more popular surface wave methods at 20 sites (Hayashi et al., 2016). We
than the FK. used the same recording geometry for both active (MASW)
and passive (MAM) surface wave methods. We acquired
Building designers widely use average S-wave velocity, the data for both MASW and MAM using a 24-channel
down to a depth of 30 m (VS30m) as a proxy for site Geometrics Geode seismograph. The receiver array
response analysis and earthquake resistance in their consisted of 24 - 4.5 Hz vertical component geophones,
building designs. In order to evaluate the VS30m quickly with a 3 m receiver interval so that the array length was 69
and inexpensively, Louie (2001) proposed the ReMi m. For MASW data acquisition, we used a 40Kg (PEG-
method, in which surface waves are recorded using a 1D 80lb) accelerated weight drop as a source. We offset the
linear array. Ambient noise data are transformed into the shots from the receiver array at 20, 10 and 3 m and stacked
phase velocity–frequency domain using a tau-p transform. 5 to 10 shots at each shot location. For MAM data
acquisition, we recorded the vertical component of ambient
Recently, researchers have been studying ambient noise in noise for ten to twenty minutes. Figure 1 shows a
seismic interferometry (Wapenaar, 2004). Conventional comparison of maximum and minimum 1/3 wavelengths
passive surface wave methods, SPAC, FK and ReMi, obtained by MAM and MASW. The MAM provide a
require spatially un-aliased data to calculate phase velocity minimum wavelength of 2 to 3 m, which is comparable
and receivers must be deployed with relatively small with the minimum wavelengths observed in the MASW of
spacing. Unlike conventional methods, seismic 1 to 2 m. However, the maximum wave length observed in
interferometry is used to calculate group velocity and the MASW is 20 to 30 m, which is much shorter than those
spatially aliased data. Researchers obtain and process data observed in MAM (40 to 100 m). Note that at only two
from relatively large spacing arrays. The group velocity, sites out of the 20 sites tested did the MASW provided a
however, is not straightforward compared with phase maximum 1/3 wavelength longer than 30 m required for a
velocity and the seismic interferometry generally requires VS30m. This implies that a penetration depth for the active
much longer record length. As a result, the method is still surface wave method might be generally insufficient for a
in the research and development phase and is not used in VS30m investigation.
routine work.
Figure 2 summarizes a relationship between array size
Horizontal to vertical spectral ratio (HVSR or H/V) of (maximum receiver separation) and the maximum
ambient noise is another widely used seismic method for wavelength obtained from SPAC with various array size
site investigation. It is generally agreed that the ambient and shapes at many different sites. The ratio of maximum
noise consists mainly of surface waves with the HVSR wave length to array size ranges from 0.7 to 10 and varies
relating to ellipticity of Rayleigh waves. In addition, there depending on the site. We can see that most data fall within
is fairly general agreement that the HVSR relates to a the range of 2 ~ 4 regardless of array size. It is generally
natural period of S-waves at the site. For the reasons given agreed that the penetration depth of surface wave methods
above, the HVSR is widely used in investigations related to is roughly 1/2 ~ 1/4 of maximum wave length. We
the site response. Unlike other surface wave methods, the therefore conclude that the penetration depth of the passive
HVSR provides neither phase nor group velocity surface wave methods using SPAC is approximately equal
information, so that estimating S-wave velocity only from to array size or maximum receiver separation.
the HVSR is generally impossible.
Applicability of irregular arrays in SPAC
Passive surface wave methods can have several huge
advantages over active methods. Some of these advantages In the presence of anisotropic ambient noise, isotropic array
include deeper penetration, the use of ambient noise which patterns such as a circle or equilateral triangle provide more
precludes the need for a noisy source, and the potential to consistent velocity measurements than anisotropic arrays
© 2017 SEG Page 5203

Figure 2: Relationship between array size (maximum

separation) and maximum wave length obtained from
Figure 1: Minimum and maximum wavelength obtained SPAC with various array size and shape at many
from active and passive surface wave methods. different sites.
such as a linear array. The anisotropic arrays, however, do the medium size array in France (Figure 4d), regardless the
provide reliable phase velocity information when ambient direction of receiver pairs.
noise propagates in many directions. Hayashi (2009)
demonstrated that a linear array can provide usable phase We have performed similar comparison at many sites and
velocities when propagation direction of ambient noise is most data does not show obvious azimuthal dependencies.
distributed over a range of at least 120 degrees. This implies that the ambient noise is generally
omnidirectional in the investigation area. This is reasonable
In order to evaluate the applicability of the anisotropic considering that the urban environment contains a plethora
array, we need to consider the azimuthal distribution of of sources that generate surface waves, including motor
ambient noise in the propagation direction. To investigate vehicles, machinery, etc. By the time the wavefields from
the azimuthal variation in the direction of propagation, we multiple sources have been multiply scattered, the resultant
analyzed ambient noise obtained from triangular or circular ambient noise field is remarkably omnidirectional. The
arrays using receiver pairs with several different uncertainty associated with anisotropic arrays can be
orientations. Figure 3 shows an example of coherencies evaluated by comparing coherencies with those obtained by
calculated from ambient noise obtained in San Francisco multi-directional arrays as shown above.
Bay Area, CA, U.S (Hayashi and Craig, 2016). Figures 3a
and 3b show medium and large size arrays for high and low Conclusions
frequency observation respectively. Figures 3c and 3d are
coherencies calculated from receiver pairs shown in Most conventional seismic exploration methods utilize
Figures 3a and 3b respectively. We can see that coherencies deterministic wave fields, and evaluating data quality is
are almost identical in the frequency range of interests, 1.5 generally straightforward. Unlike the conventional methods,
to 5 Hz in the medium size array (Figure 3c) and 0.2 to 0.5 passive surface wave methods utilize stochastic wave fields
Hz in the large size array (Figure 3d), regardless the and evaluating the quality or reliability of data is not
direction of receiver pairs. intuitive. Preliminary processing in the field is required to
evaluate the data quality of passive surface wave methods
Figure 4 shows a similar comparison for Italy (4a and 4c) since raw data obtained from these methods looks noisy.
and France (4b and 4d). Data were obtained in the Applicability and reliability of the method depend very
international blind tests of surface wave methods (Garofalo much on site conditions, and it is dangerous to rely on only
et al., 2016). We can see that coherencies are almost one method. Combining several different methods, for
identical in the frequency range of interests, 1.5 to 6 Hz in example active and passive methods, is generally
the small size array in Italy (Figure 4c) and 0.5 to 2 Hz in recommended. Storing field examples and sharing
empirical knowledge are also important.
© 2017 SEG Page 5204

Figure 3: Example of coherencies calculated from linear arrays with different directions. a) Array configuration for high
frequency measurements. b) Array configuration for low frequency measurements. c) Coherencies calculated from a
medium size array (a). d) Coherencies calculated from a large size array (b).
Figure 4: Example of coherencies calculated from linear arrays with different directions. a) Array configuration for high
frequency measurements. b) Array configuration for low frequency measurements. c) Coherencies calculated from a small
size array (a). d) Coherencies calculated from a medium size array (b).
© 2017 SEG Page 5205

EDITED REFERENCES
REFERENCES
Aki, K., 1957, Space and time spectra of stationary stochastic waves, with special reference to
microtremors: Bulletin of the Earthquake Research Institute, 35, 415–456.
Garofalo, F., S. Foti, F. Hollender, P. Y. Bard, C. Cornou, B. R. Cox, M. Ohrnberger, D. Sicilia, M.
Asten, G. Di Giulio, T. Forbriger, B. Guillier, K. Hayashi, A. Martin, S. Matsushima, D.
Mercerat, V. Poggi, and H. Yamanaka, 2016, InterPACIFIC project: Comparison of invasive and
non-invasive methods for seismic site characterization. Part I: Intra-comparison of surface wave
methods: Soil Dynamics and Earthquake Engineering, 82, 222–240,
http://doi.org/10.1016/j.soildyn.2015.12.010.
Hayashi, K., 2009, Effect of array shape on the spatial auto-correlation analysis of micro-tremor array
measurements: Proceedings of the Symposium on the Application of Geophysics to Engineering
and Environmental Problems, 616–625, https://doi.org/10.4133/1.3176749.
Hayashi, K., R. Cakir, and T. J. Walsh, 2016, Comparison of dispersion curves obtained by active and
passive surface wave methods: Examples from seismic site characterization surveys for school
seismic safety evaluations in Thurston County, WA: Symposium on the Application of
Geophysics to Engineering and Environmental Problems, 475–483,
https://doi.org/10.4133/sageep.29-081.
Hayashi, K., and M. Craig, 2016, S-wave velocity measurement and the effect of basin geometry on site
response, east San Francisco Bay area, California, USA: Physics and Chemistry of the Earth Parts
A/B/C, 98, 49–61, http://doi.org/10.1016/j.pce.2016.07.001.
Hayashi, K., and H. Suzuki, 2004, CMP cross-correlation analysis of multi-channel surface-wave data:
Exploration Geophysics, 35, 7–13, http://doi.org/10.1071/EG04007.
Louie, J. M., 2001, Faster, better: Shear-wave velocity to 100 meters depth from refraction microtremor
arrays: Bulletin of the Seismological Society of America, 91, 347–364,
http://doi.org/10.1785/0120000098.
Miller, R. D., J. Xia, C. B. Park, and J. M. Ivanov, 1999, Multichannel analysis of surface waves to map
bedrock: The Leading Edge, 18, 1392–1396, http://doi.org/10.1190/1.1438226.
Nazarian, S., K. H. Stokoe, and W. R. Hudson, 1983, Use of spectral analysis of surface waves method
for determination of moduli and thickness of pavement system: Transportation Research Record,
930, 38–45.
Okada, H., and K. Suto, 2003, The microtremor survey method: SEG, Monograph Series 12,
https://doi.org/10.1190/1.9781560801740.
Park, C. B., R. D. Miller, and J. Xia, 1999a, Multimodal analysis of high frequency surface waves:
Proceedings of the Symposium on the Application of Geophysics to Engineering and
Environmental Problems, 115–121.
Park, C. B., R. D. Miller, and J. Xia, 1999b, Multichannel analysis of surface waves: Geophysics, 64,
800–808, http://doi.org/10.1190/1.1444590.
Wapenaar, K., 2004, Retrieving the elastodynamic Green’s function of an arbitrary inhomogeneous
medium by cross correlation: Physical Review Letters, 93, Article 254301,
http://doi.org/10.1103/PhysRevLett.93.254301.
Xia, J., R. D. Miller, and C. B. Park, 1999, Configuration of near surface shear wave velocity by inverting
surface wave: Proceedings of the Symposium on the Application of Geophysics to Engineering
and Environmental Problems, 95–104, https://doi.org/10.4133/1.2922698.
© 2017 SEG Page 5206

Surface-wave methods for anomaly detection, a review
J. Tyler Schwenk*, Primal Innovation and Steven D. Sloan, US ARMY ERDC
Summary MASW approach to 3D. Ivanov et al. (2013) used passive

MASW techniques to image the anomalous velocity regime
Focusing on tunnels, we review several surface-wave method over an abandoned salt injection well. Samyn et al. (2014)
(SWM) innovations used in the hunt for near-surface performed anomaly and hazard identification along a dyke
anomalies. We primarily examine a recent publication in system affected by karst dissolution using MASW velocity
Geophysics that detailed several advancements of the analysis. Lastly, Samyn et al. (2013) applied cross correlation
classical backscatter analysis of surface waves (BASW) and coda wave interferometry theory through a linear move
technique that has proven reliable through multiple field out operation to map the Vs perturbations associated with
investigations in locating anomalies. There, 3D visualization, near-surface anomalies.
modal separation, simple linear correlation, and amplitude-
envelope analysis was integrated into the BASW scheme with In the presence of lateral discontinuities, an incident wave
positive results. This multifaceted SWM approach provided will create a scattered wavefield (Herman et al., 2000; Ernst
coincident data sets that increase confidence, and reduce false et al., 2002; Strobbia et al., 2014; Halliday et al., 2015).
positives, when trying to detect and localize subsurface MASW images anomalies through an apparent change in
discontinuities. Our review of this, and other researchers’ velocity. SWMs may also image the scattered wavefield
works, suggests SWMs offer a substantial asset to anomaly- directly or attempt to do so indirectly through attenuation
detection studies. analysis. Moreover, we can distinguish the forward-
propagating and ‘backscattered’ energy as those wavefields
Introduction traveling away and towards the source, respectively. Many
techniques lump both subsets together during processing,
Engineering and environmental geophysical investigations while others seek to enhance or better define them
encounter many situations where anomaly-detection is individually.
paramount. Dissolution phenomena can create hazardous
conditions with the failure of mining and drilling structures. While adversely affected by both coherent and non-coherent
Similarly, the geomorphology associated with karst noise, researchers have found success in anomaly detection
environments affects many population centers. Construction using surface-wave attenuation analysis. Nasseri-
and infrastructure concerns encompass another subset with Moghaddam et al. (2005) proposed the attenuation analysis
the risk associated with unstable structural foundations, earth- of Rayleigh waves (AARW) technique to image voids.
retaining structures, dams, and levees. Lastly, clandestine Bergamo and Socco (2014) successfully applied, and further
tunneling affects both civilian and law-enforcement/military developed, the AARW method to sub-vertical fracture zones
centers worldwide. Whether in the search for voids, tunnels, with dual-direction acquisition and processing. Although
or weak-strength zones, the geophysical community has a based on attenuation, both AARW studies note magnification
clear objective to detecting anomalies across a wide range of is apparent and coupled to the constructive and destructive
environments and applications. interaction of the scattered wavefield.
Surface-wave methods (SWM) have long been used to assess The advancements found in the Schwenk et al. (2016) are
the shear-wave velocity (Vs) of the subsurface. Today, many reviewed. We use the backscatter analysis of surface waves
site characterization studies use a distributed geophone array (BASW) technique to image void-like anomalies in the
during acquisition following the multi-channel analysis of subsurface (Sloan et al., 2015). Mode-specific f-k muting and
surface waves (MASW) technique (Miller et al., 1999) and frequency-variant linear move out(s) (FV-LMO; Park et al.,
similar variations in 2D and 3D (Louie, 2001; Park and 2002; Park et al., 1998) suggest we may image scatterers
Taylor, 2010; Wang et al., 2015). Although anomalies can be discretely due to their multi-mode character. Simplistic
interpreted from MASW velocity results, the multi-channel correlation of multi-mode corrected BASW results allows for
dispersion-curve transforms smear the velocity field across a quantitative interpretation of the BASW images. Three-
the array and ultimately reduce the resolution power of the dimensional visualization routines allow for rapid
method limiting the size of anomalies that can be imaged interpretation of site characteristics within an instantaneous-
(e.g., Boiero and Socco, 2011). amplitude cube. Incorporating instantaneous-amplitude into
the complex BASW technique provides a more robust
Regardless of the questions on resolution, several studies implementation with reduced sensitivity to velocity
have had success imaging Vs anomalies. Park and Taylor correction. Although our imaging method is designed to
(2010) investigated a sinkhole by expanding the standard 2D enhance backscatters, our results suggest the forward-going
© 2017 SEG Page 5207

Surface-wave anomaly detection: a review
surface wave may be an untapped resource for anomaly character. We use the Hilbert transform (Taner et al., 1979)
detection using BASW. and sort the instantaneous-amplitude data into a 3D lattice
(i.e., common offset vs. common receiver vs. time). We refer
Correlation analysis, mode separation and multi-mode BASW to this 3D visualization as the amplitude cube.
imaging, complex BASW, and 3D amplitude-cube
visualization are the major contributions of Schwenk et al. Surgical f-k muting of the forward-propagating surface waves
(2016). A multi-pronged SWM investigation allows for sought modal separation of the fundamental and HM
confident localization and detection of subsurface anomalies. signatures while minimizing the effects of spectral smearing,
ringing, and signal loss. BASW images overlay
Methods instantaneous-amplitude transforms to emphasize dynamic
changes in trace bandwidth and energy.
Each method is described with limited detail; for more
information, we suggest an in-depth review of Schwenk et al. Lastly, we incorporate the Hilbert transform into a complex
(2016). BASW processing scheme – the trace envelope is computed
from raw-data traces before following the standard BASW
We use the multi-channel analysis of surface waves (MASW) procedures with the noted exception of f-k filtering. The
survey technique (Miller et al., 1999) and the high-resolution method also uses a singular correction velocity that is the
linear radon transform to analyze Rayleigh-wave dispersion average Vs associated with a suspected anomaly’s depth.
(Luo et al., 2008). Dispersion curves are picked for apparent
fundamental and higher-mode (HM) curves. The BASW Data sets were collected with a land-streamer array of 24 4.5-
method uses these curves to produce mode-specific FV-LMO Hz vertical geophones spaced every 1.2 m (4 feet) with 2.4 m
corrections that flatten the forward-going surface wave source rolls. The source was an accelerated weight drop that
placing the backscattered loci along the offset axis. A offers high-energy, broadband, and highly repeatable impacts
common receiver stack results in the BASW image. Figure 1 that saves man-power in respect to a sledgehammer. All lines
is an example of BASW images derived from a modeled void. were orientated roughly perpendicular to the long axis of the
tunnels. Quality comparisons between steamer and fixed
spreads show negligible differences in signal quality (e.g., van
der Veen et al., 2001). The results are cropped up to one
streamer length at their edges to reduce low-fold artifacts.
Results
Correlation
We apply the correlation processing to a purpose-built 1.2 m

x 1.5 m, 9.1-m deep tunnel, with a 36.6-m source offset.
Figure 2a depicts the fundamental mode BASW image. The
three correlation products include: fundamental and higher-
mode BASW image correlation (Figure 2b); correlation of the
frequency spectra of the modal BASW images (Figure 2c);
and correlation of the directional derivatives across offset of
those frequency spectra (Figure 2d). With each successive
correlation product, the interpretation of the BASW image
Figure 1: Classical BASW results for a 1.2 m x 1.5 m, 9.1-m deep
modeled void denoted by the bull’s eye (after Schwenk et al.,
may become more confident.
2016). Both BASW images use the same f-k filter but apply a)
fundamental and b) HM FV-LMOs. Both images properly image Amplitude Cube
the void’s location.
Here, we investigate a purpose-build 1.25 m x 1.25 m, 3-m
Our correlation processing uses the Pearson product-moment deep tunnel, with a 14.6-m source offset. The tunnel anomaly
correlation procedure to examine the linear correlation (r) of is clearly visible on the amplitude cube at station 1018 (Figure
offset- and time-consistent samples of multi-mode BASW 3a). Interestingly, enhancing the backscattered wavefield
images (i.e., fundamental and HM corrections) on a trace-by- using a forward f-k filter destroys the scattered signature
trace basis (Pearson, 1895; Dubrule, 2003). The square of the (Figure 3b). Combining forward- and reverse-acquisition
correlation coefficient (r2) is then graphed across offset as a schemes increases the coherence of the amplitude horizons
function of the midpoint of a ‘bin’ of several traces. resulting in a semi-symmetrical tunnel signature (Figure 3c).
The important distinction of these findings is that we are
Scanning through common-offset and common-receiver imaging a forward-scattering phenomenon that is lost with
stacks allows for quick inference of lateral variance in seismic muting of the forward wavefield.
© 2017 SEG Page 5208

mode FV-LMOs give discrete backscatter

signatures at the tunnel location. Integrating
forward- and reverse-direction acquisition
illustrates the close relationship between the
BASW image and the amplitude-cube scheme
with a semi-symmetrical signature (Figure 4d).
Figure 4d enhances the forward-propagating

anomaly with dual-direction lines – the same
technique can be applied for the backscatter, but
is not shown here.
Complex BASW
The last method incorporates instantaneous-

amplitude analysis into a ‘complex’ BASW
routine. The 3-m tunnel discussed previously is
compared to a 22-m deep clandestine tunnel.
Both sites’ processing combines dual-direction
source rolls to image semi-symmetrical
signatures and increase S/N. The clandestine
tunnel data was acquired with a 1-m geophone
spacing, a 1-m source roll, and a fixed 72-
geophone spread – the tunnel is approximately
1 m x 1 m in cross-section.
Using a shorter source offset of 7.3 m, the 3-m

deep tunnel exhibits a clearly defined anomaly
(Figure 5a). Our next example is a clandestine,
approximately 1 m x 1 m, 22-m deep tunnel
Figure 2 (Left): a) Fundamental BASW image of a 9.1-m deep tunnel located at station (Figure 5b). Both sites’ complex BASW images
1158. b) Fundamental and HM BASW image correlation diagram. c) Correlation produced an inverted ‘V’, centered on the
diagram using the frequency spectra of the modal BASW images. d) Correlation tunnels, this mimics the amplitude-cube trends
diagram using the directional derivatives of the modal BASW frequency spectra (after across offset slices and the slanted signatures of
Schwenk et al., 2016). classical BASW imaging.
Figure 3 (Right): Amplitude cube a) of a field data set over a 3-m deep tunnel with a
clear high-amplitude signature; b) after applying a forward f-k filter, the forward-scatter Discussion
void signature is destroyed; c) combining forward- and reverse-acquisition schemes
with a semi-symmetrical signature centered on the void location. Artifacts are apparent The methods included in this review offer
at the edges of the common-offset slices; they are a product of poor receiver coverage. substantial advancements to the classical
The black cuboid represents the tunnel location (after Schwenk et al., 2016). BASW method. The amplitude cube allows for
.
rapid interpretation of anomalies and general
Mode separation & Multi-mode BASW velocity trends across a site. Correlation analysis moves the
subjective interpretation of BASW images towards a more
Revisiting the 3-m tunnel with a shorter source offset of 7.3 quantitative assessment. Modal separation produces
m, mode-specific FV-LMO corrections and f-k filtering coincident data sets that ultimately reduce false positives and
enhance individual scatter events (Figure 4). Depending on increase confidence in points of interest. The complex BASW
the FV-LMO, the scatter loci deviate across several stations process improves method robustness by eliminating the
(Figure 4a-4d, 1018-1021). Although small adjustments to ambiguity of dispersion interpretation, with the use of a single
the picked dispersion curve(s) would align all the results to correction velocity, and the removal of f-k filtering.
the correct location, we wanted to show the effects of real-
world interpretations. We haven’t promoted a strict comparison of every method
across each site because we cannot always depend on
The classical approach enhances the ‘fundamental’ unambiguous reinforcement and an individual method’s
backscatter with a fundamental f-k filter and FV-LMO success. The correlation procedure will fail without consistent
(Figure 4a). As with the amplitude cube, there is a dominant HM excitation. A chaotic velocity regime may require a more
forward-propagating tunnel signature – it was enhanced with complicated velocity function across a line. Depth penetration
a HM f-k filter and FV-LMO (Figure 4b). An additional will not only effect what modes and frequency bands interact
fundamental-mode mute decimates the scattered anomaly and with a subsurface anomaly, but we also require increasingly
enhances the ‘HM’ backscatter (Figure 4c). Using multi- larger-energy sources to push to lower frequencies and
© 2017 SEG Page 5209

greater depths. Where one procedure may fail, the use of

multiple methods encourages the positive identification of
anomalies.
Mode-separation is essential to defining the various coherent

signals produced from seismic anomalies and lateral
discontinuities. The BASW method approaches this with FV-

LMOs, f-k and bandpass filtering. Quantitative amplitude
analysis (i.e., attenuation and magnification) should take this
into consideration. The constructive and destructive
interference inherent in modal superposition may make
attenuation calculations more related to mode excitation and
interference patterns rather than anomaly signatures.
Recent history has clandestine tunnels averaging

approximately 1 m x 1 m. Dissolution cavities and weak
shear-strength anomalies may be smaller or many times larger
than this. Fault studies may investigate sharp lateral
discontinuities, but also generally encompass bigger cross-
sectional areas. Having focused on relatively small targets,
we are confident that these procedures may also be applied to
larger-scale anomalies.
Figure 4: Wavefield-separation examples using a) a classical
Conclusion fundamental f-k mute and FV-LMO; b) a HM f-k mute and FV-
LMO; c) both fundamental and HM f-k mute(s) with a HM FV-
Our methods can be summarized as operations that image LMO; d) forward and reverse lines merged into one routine with
perturbations in seismic-velocity with emphasis on surface- a HM f-k mute and a FV-LMO. The white vertical line is the
wave modes. Future research should develop procedures that approximate tunnel location, station 1018 (after Schwenk et al.,
better separate the mode- and frequency-dependent responses 2016).
of seismic anomalies. Surface waves will also play a role in
the developing field of full-waveform inversion (FWI).
Specifically, FWI often requires high-confidence velocity
models to accurately converge to stable and accurate
solutions. Nevertheless, preliminary studies show great
promise in the use of FWI for imaging anomalies and research
has emphasized the importance of the backscattered
wavefield in doing so.
The methods found in this review have proven reliable in

localizing void-like bodies across multiple field sites and
anomaly depths. Moreover, our results suggest the forward-
going surface wave may be an untapped resource for anomaly
detection using BASW. Correlation, amplitude-cube
visualization, mode separation, and complex BASW imaging
represent advancements of the BASW method and may prove
valuable for perimeter defense, geohazard assessment, and
site characterization investigations.
Acknowledgments
Many thanks go to deployed personnel for their collection of

data outside the continental US. As this is a review of
previously published work, we also acknowledge the co-
authors and work completed in the production of Schwenk et Figure 5: Two field studies are compared using the complex
al. (2016). BASW method for: a) the previous 3-m deep tunnel; b) a
clandestine 22-m deep tunnel (after Schwenk et al., 2016). The
white circles mark the tunnel locations. The black rectangles point
to areas of both images that exhibit remarkably similar amplitude
character.
© 2017 SEG Page 5210

EDITED REFERENCES
REFERENCES
Bergamo, P., and L. V. Socco, 2014, Detection of sharp lateral discontinuities through the analysis of
surface-wave propagation: Geophysics, 79, no. 4, EN77–EN90,
Boiero, D, and L. V. Socco, 2011, The meaning of surface wave dispersion curves in weakly laterally
varying structures: Near Surface Geophysics, 9, 561–570. http://dx.doi.org/10.3997/1873-
0604.2011042.
Dubrule, O., 2003, Geostatistics for seismic data integration in earth models: SEG Distinguished
Instructor Series No. 6, SEG and EAGE, 273.
Ernst, F. E., G. C. Herman, and A. Ditzel, 2002, Removal of scattered guided waves from seismic data:
Geophysics, 67, 1240–1248, http://dx.doi.org/10.1190/1.1500386.
Halliday, D., P. Bilsby, L. West, E. Kragh, and J. Quigley, 2015, Scattered ground-roll attenuation using
model-driven interferometry: Geophysical Prospecting, 63, 116–132,
http://dx.doi.org/10.1111/1365-2478.12165.
Herman, G. C., P. A. Milligan, R. J. Huggins, and J. W. Rector, 2000, Imaging shallow objects and
heterogeneities with scattered guided waves: Geophysics, 65, 247–252,
http://dx.doi.org/10.1190/1.1444715.
Ivanov, J., B. Leitner, W. Shefchik, J. T. Schwenk, and S. Peterie, 2013, Evaluating hazards at salt cavern
sites using multichannel analysis of surface waves: The Leading Edge, 32, 298–305,
http://dx.doi.org/10.1190/tle32030298.1.
Louie, J. N., 2001, Faster, better: Shear-wave velocity to 100 meters depth from refraction microtremor
arrays: Bulletin of the Seismological Society of America, 91, 347–364,
http://dx.doi.org/10.1785/0120000098.
Luo, Y., J. Xia, R. D. Miller, Y. Xu, J. Liu, and Q. Liu, 2008, Rayleigh-wave dispersive energy imaging
using a high-resolution linear Radon transform: Pure and Applied Geophysics, 165, 903–922,
http://dx.doi.org/10.1007/s00024-008-0338-4.
Miller, R. D., J. Xia, C. B. Park, and J. M. Ivanov, 1999, Multichannel analysis of surfaces waves to map
bedrock: The Leading Edge, 18, 1392–1396, http://dx.doi.org/10.1190/1.1438226.
Nasseri-Moghaddam, A., G. Cascante, and J. Hutchinson, 2006, A new quantitative procedure to
determine the location and embedment depth of a void using surface waves: Journal of
Park, C.B., and C. Taylor, 2010, 3D MASW characterization of sinkhole: A pilot study at USF Geology
Park, Tampa, Fl: Proceedings of the symposium on the application of geophysics to engineering
and environmental problems, 498–507.
Park, C. B., R. D. Miller, and J. Ivanov, 2002, Filtering surface waves: Proceedings from the Symposium
for the Application of Geophysics to Environmental and Engineering Problems, SEI9.
Park, C. B., R. D. Miller, and J. Xia, 1998, Ground roll as a tool to image near-surface anomaly: 68th
Annual International Meeting, SEG, Expanded Abstracts, 874–877,
http://dx.doi.org/10.1190/1.1820627.
Pearson, K., 1895, Contributions to the mathematical theory of evolution. II. Skew variation in
homogeneous material: Philosophical Transactions of the Royal Society of London, Series A,
186, 343–414, http://dx.doi.org/10.1098/rsta.1895.0010.
© 2017 SEG Page 5211

Samyn, K., F. Mathieu, A. Bitri, A. Nachbauer, and L. Closset, 2014, Integrated geophysical approach in
assessing karst presence and sinkhole susceptibility along flood-protection dykes of the Loire
River, Orleans, France: Engineering Geology, 183, 170–184,
https://doi.org/10.1016/j.enggeo.2014.10.013.
Samyn, K., A. Bitri, and G. Grandjean, 2013, Imaging a near-surface feature using cross-correlation
analysis of multi-channel surface wave data: Near Surface Geophysics, 11, 1–10,
https://doi.org/10.3997/1873-0604.2012007.
Schwenk, J. T., S. D. Sloan, J. Ivanov, and R. Miller, 2016, Surface-wave methods for anomaly detection:
Geophysics, 81, EN29–E42, http://dx.doi.org/10.1190/geo2015-0356.1.
Sloan, S. D., S. L. Peterie, R. D. Miller, J. Ivanov, J. T. Schwenk, and J. R. McKenna, 2015, Detecting
clandestine tunnels using near-surface seismic techniques: Geophysics, 80, no. 5, EN127–EN135,
http://dx.doi.org/10.1190/geo2014-0529.1
Strobbia, C., A. Zarkhidze, F. Ibrahim, and R. May, 2014, Model-based attenuation for scattered
dispersive waves: Geophysical Prospecting, 62, 1143–1161, http://dx.doi.org/10.1111/1365-
2478.12118.
Taner, M. T., F. Koehler, and R. E. Sheriff, 1979, Complex seismic trace analysis: Geophysics, 44, 1041–
1063, http://dx.doi.org/10.1190/1.1440994.
van der Veen, M., R. Spitzer, A. G. Green, and P. Wild, 2001, Design and application of a towed land-
streamer system for cost-effective 2-D and pseudo-3-D shallow seismic data acquisition:
Geophysics, 66, 482–500, http://dx.doi.org/10.1190/1.1444939.
Wang, L., Y. Xu, and Y. Luo, 2015, Numerical investigation of 3D multichannel analysis of surface wave
method: Journal of Applied Geophysics, 119, 156–169,
https://doi.org/10.1016/j.jappgeo.2015.05.018
© 2017 SEG Page 5212

Compaction Evaluation by MASW Surveys (CEMS)
Choon Park*, Park Seismic LLC, and Jason Richter, Minnesota Department of Transportation (MnDOT)
Abstract defined by a material's density () and the two seismic

velocities (or by Poisson's ratio, ) of Vp (P-wave) and Vs
The purpose of the compaction during a road construction (S-wave):
is to achieve a necessary stiffness level. Therefore, the
E = 2Vs2(1+) (1)
compaction evaluation process can be regarded as an in-situ
µ = Vs2 (2)
shear-wave velocity (Vs) measurement for top, at most,
1-m thickness of roadbeds. Minnesota Department of The two defining equations indicate the heaviest
Transportation (MnDOT) recently launched a feasibility dependency of both moduli on Vs. This is why seismic
study to use a seismic surface-wave (MASW) method for shear-wave velocity (Vs), the final product from an MASW
this purpose. The first field test, conducted at a 153-m-long survey (Park et al., 1999), is often used as a direct indicator
test site, used a set of four parallel land streamers (12- of a material's stiffness.
channel per streamer) with 1-m geophone spacing and
1.2-m streamer separation. Five surveys were conducted at MnDOT recently recognized the potential utility of MASW
five different compaction stages during a full-depth surveys in compaction evaluation during road construction
reclamation (FDR) road construction. Analysis results of and launched a feasibility field study to tap into its
cross sections clearly showed velocity (Vs) variations for, effectiveness and move toward the goal of making it a
approximately, the top 0.3-m thickness between different routine production method. The pilot study consisted of a
stages. In addition, depth-slice (DS) maps created from series of multiple (five) MASW surveys performed in the
four parallel cross sections delineated velocity variations same area during a full-depth-reclamation (FDR) road
not only between different stages, but also between construction; a 153-m long segment on TH56 South
different surface locations. Frequency limitations in approximately 5 miles north of Kenyon, MN (Figure1).
surface waves measured by using low-frequency (4.5 Hz)
geophones and relatively long (11 m) receiver array for
depth of interest (< 1 m) made overall velocities
underestimated. Future adjustment in acquisition system
and geometry will significantly increase the resolution of
the approach.
Introduction
The main purpose of the compaction process applied at

various stages of road construction is to achieve the level of
stiffness necessary to sustain expected load stress over the
entire construction area. In this sense, the compaction
evaluation can be regarded identical to in-situ stiffness
measurement of road materials.
Stiffness of a material is defined as a measure of resistance

Figure 1. Location map and quadruple land streamers used
to deformation (Sheriff, 2002) and ultimately related to
material's elastic moduli that describe the material's to acquire MASW data at a test site near Kenyon, MN.
behavior under stress. Among the three primary types of Road Construction by Full-Depth Reclamation (FDR)
moduli—Young's (E), shear (µ), and bulk ()—the first
two (E and µ) are most commonly used because of what The full-depth reclamation (FDR) rebuilds old worn-out
they represent. Young's modulus (E) simply dictates the asphalt pavements by recycling the existing roadway. The
deformation tendency along the axis of stress, whereas the old asphalt and base materials are pulverized, mixed with
shear modulus (µ) indicates the tendency of shape stabilizing agent (e.g., cement, emulsified asphalt, or
deformation (i.e., shearing). In reality, deformation always asphalt binder, etc.), and compacted to produce a stabilized
accompanies both transverse and longitudinal changes only base. The new pavement layer of asphalt is then laid on top
at a different ratio. In this sense, the most comprehensive of the stabilized base. The overall procedure can be
and accurate definition of stiffness should include both divided into three stages; (1) the pre-grind (PG) stage in
moduli of E and µ. According to the theory of elasticity which existing old pavement and base layers are
(Sheriff and Geldart, 1982), these two moduli can be pulverized, (2) the material is then re-ground and cement,
© 2017 SEG Page 5213

emulsified asphalt, or an asphalt binder is added and this depth of 2.0 m (Figure 2a). The modeled AM0 curve is
layer is compacted to produce a stabilized FDR (SFDR), displayed (in black) in Figure 2b in comparison to three
subsequently after curing, (3) layer(s) of hot-mix-asphalt other AM0 curves that are modeled after changing
(HMA) is placed. During and between each stage, it is (increasing) the velocity (Vs) by 30% in each of the three
important to ensure the necessary level of stiffness is layers; i.e., Vs1, Vs2, and Vs3, respectively. The upper
achieved over the entire area of construction. Standard limit of modeled frequency is 1000 Hz where the AM0
penetration test (SPT) is traditionally conducted for this curves approach the asymptotic surface-wave velocity of
purpose at selected locations, while more recently top layer by more than 98%. The curve comparison shown
Intelligent Compaction (IC) techniques are used. MASW in Figure 2b indicates that velocity (Vs) changes in the
surveys are proposed as a robust QA/QC means that can three layers result in phase velocity changes mostly at those
provide distribution of stiffness in more technically frequencies higher than 100 Hz (top base), 30-200 Hz
appropriate and spatially continuous form than other (subgrade), and lower than 20 Hz (bedrock), respectively.
approaches can provide. It is also obvious that the greatest overall change occurs
when the top layer changes its velocity, proving the highest
Data Acquisition sensitivity despite the smallest thickness. These results
therefore indicate that, as far as the highest measured
Total five (5) MASW surveys were conducted at the same frequency (fmax) exceeds 100 Hz (and the lowest frequency
place during the FDR; one during the pre-grind stage (PG), is lower than 20 Hz), velocity change in any of these three
two during the stabilized FDR stage (SFDR and SFDRb), layers will be detected. However, velocity for top layer
and two during the final HMA stage (HMA1stLift and (Vs1) will be underestimated while fmax remains lower than
HMA2ndLift). The surveys took place during July and 1000 Hz. Figure 2b indicates that, as fmax increases and
August, 2013, by using the existing seismic acquisition approaches 1000 Hz, the degree of underestimation will
system built and used by MnDOT Materials with a minimal decrease, while the sensitivity in detecting velocity (Vs1)
modification, a seismic system that employed low- change will increase.
frequency geophones (4.5-Hz) for receivers and was
originally built for subsurface investigation at deeper
depths of soil and bedrock (e.g., 1-30 m) than the current
depth of interest (e.g., 0-2 m). The system consisted of 48-
channel acquisition with quadruple land streamers (12
channels/streamer) placed parallel and separated by 1.2-m
(Figure 1). A weight-drop source (WD/SASS) generated
surface waves 2-m ahead of the closest geophones from the
transverse center of the streamers. One impact was
delivered to generate one 48-channel field record at one
location, and this source-receiver (SR) configuration
moved by 1 m each time to produce a total of 154 field
records per survey ensuring the coverage of 153-m (500-ft-
long) segment of the test site.
Resolution Analysis
Considering the acquisition geometry of receiver array

length (L=11 m) and source offset (X1=2 m), the maximum
investigation depth (Zmax) is expected to be about 5 m.
However, most stiffness changes that will occur during the
FDR construction are confined within about the upper
0.3-m thickness. Therefore, it is worth conducting an
analysis of how the stiffness variations in this relatively
thin layer, as well as other underlying layers, influence the
dispersion curve measurement. This analysis is performed Figure 2. (a) Layer model used to generate (b) theoretical
by modeling an apparent-mode (AM0) dispersion curve apparent-mode (AM0) dispersion curves. The original
(Gucunski and Woods, 1992) for a velocity (Vs) model that curve (in black) is displayed in comparison to other curves
can represent a typical road base (Vs1=300 m/sec) of 0.3-m generated by increasing velocity by 30% for top (Vs1),
thickness overlying a subgrade (Vs2=150 m/sec) followed subgrade (Vs2), and weathered bedrock (Vs3) layers.
by weathered bedrock (Vs3=500 m/sec) at an arbitrary
© 2017 SEG Page 5214

Data Analysis and Results stages of FDR road construction as well as between
different surface locations. This is an unprecedented
The acquired data set from each survey was split into four approach that deals with the most important property (i.e.,
(4) subsets of individual lines (1-4) (Figure 1) stiffness) of road materials through one of the most
corresponding to each land streamer of 12-channel fundamental scientific approaches; i.e., through seismic-
acquisition during the pre-processing step. Each subset wave propagation.
then went through the normal MASW data analysis
sequence to generate a 2-D velocity (Vs) cross section per Considering the possible range of shear-wave velocities for
line. base materials (e.g., 200-500 m/sec) and bituminous
pavement (e.g., 1,000-2,000 m/sec), and also possible
Different offline offsets of source locations for each line thickness ranges (e.g., 0.1-0.5 m for base, and 0.05-0.30 m
were accounted for during dispersion analysis. Then, for pavement), the optimum frequency ranges necessary for
dispersion images were generated for a frequency range of absolute evaluation of each layer's velocity are calculated,
1-1000 Hz (0.5-Hz increment) and a phase-velocity range approximately, as 500-5,000 Hz for the base and
of 10-1500 m/sec (5 m/sec increment). Dispersion curves 5,000-30,000 Hz for the pavement, respectively (Ryden et
were next extracted from these images in an approximate al., 2004). Therefore, the results (15-200 Hz) from the five
common frequency range of 15-200 Hz. A set of dispersion (5) field surveys represent underestimated velocities for
curves for each line was then used for inversion analysis to base and pavement layers, especially for the pavement. A
produce a 2-D velocity (Vs) map of 2-m depth, which was significant improvement in results is expected if a smaller
set intentionally smaller than the optimum depth (e.g., 5 m) geophone spacing (e.g., 0.5 m) is used. This reduced
to increase the resolution at shallower depths (e.g., ≤ 1 m). receiver spacing (dx) by itself will increase the maximum
A 15-layer earth model of varying thicknesses was used frequency of measured dispersion curves by, for example,
during the inversion. In this way, four (4) cross section Vs two times if dx is reduced by half. In addition, switching to
maps were produced from each survey for the four (1-4) higher-frequency geophones (e.g., 40-Hz phones or 100-Hz
parallel lines. The Vs cross sections for line 1 are phones) will increase overall recording sensitivity at higher
displayed in Figure 3 for all five (5) surveys. frequencies (e.g., 100-1,000 Hz) and can ultimately
improve the analysis resolution. However, for absolute
Considering the four (4) lines of 2-D Vs maps being velocity (Vs) evaluation of base and pavement layers,
located side by side with an even spacing (1.2 m) between accelerometers have to be used that can record surface
them, it is possible to construct depth-slice (DS) maps by waves up to 50,000 Hz (50 KHz).
combining Vs data sets from all four lines. In this way, a
DS map for 0.0-0.30 m depth range was created for each The current data analysis sequence requires an operator's
survey from four (4) lines of 2-D Vs maps. These velocity continuous involvement at several different stages of
(Vs) DS maps are displayed in Figure 4 for all five (5) processing. Most of these steps will eventually be fully
surveys. automated within the analysis software, eliminating the
need for operator intervention. This fully-automated
Then, these DS maps are converted to Young's (E) and software will lead to a complete system in the field that will
shear () moduli values by using the two equations of (1) produce cross sections and depth-slice (DS) maps of
and (2). Corresponding DS maps are displayed in Figures stiffness in real-time mode as field survey proceeds.
5a and 5b, respectively. A constant density () of 2000
kg/m3 and also a constant Poisson's ratio () of 0.4 were
used during the conversion. Acknowledgments
Conclusions and Recommendations We thank Rebecca Embacher of the MnDOT Advanced

Materials and Technology Unit for providing project
Although geophones are normally used to investigate a coordination during construction. We also thank Joe
greater depth range (e.g., 0-30 m), they were used in this Hudak and Dan Mattison of the MnDOT Geotechnical
study mainly to investigate stiffness variations in top about Engineering Section for their assistance with data
0.5-m road beds. Despite the reduced resolution in acquisition.
measurement due to frequency limitations in measured
data, it seems that results successfully showed relative
variations in stiffness that were expected between different
© 2017 SEG Page 5215

Figure 3. Velocity (Vs) cross sections of line 1 from the Figure 4. Velocity (Vs) depth-slice (DS) maps from the
five (5) MASW surveys; i.e., PG, SFDR, SFDRb, five (5) MASW surveys for 0.0-0.3 m depth. Each DS map
HMA1stLift, and HMA2ndLift (from top to bottom). is constructed from four (4) parallel cross sections.
Figure 5. Velocity (Vs) depth-slice (DS) maps in Figure 4 are converted to (a) Young's (E) and (b) shear () moduli maps by
using constant values of density () and Poisson's ratio ().
© 2017 SEG Page 5216

EDITED REFERENCES
REFERENCES
Gucunski, N., and R. D. Woods, 1992, Numerical simulation of the SASW test: Soil Dynamics and
Earthquake Engineering, 11, 213–227, https://doi.org/10.1016/0267-7261(92)90036-D.
Park, C. B., R. D. Miller, and J. Xia, 1999, Multichannel analysis of surface waves: Geophysics, 64, 800–
808, https://doi.org/10.1190/1.1444590.
Ryden, N., C. B. Park, P. Ulriksen, and R. D. Miller, 2004, Multimodal approach to seismic pavement
testing: Journal of Geotechnical and Geoenvironmental Engineering, 130, 636–645,
https://doi.org/10.1061/(ASCE)1090-0241(2004)130:6(636).
Sheriff, R. E., 2002, Encyclopedic dictionary of applied geophysics, 4th ed.: SEG Geophysical Reference
Series No. 13, SEG, p. 429.
Sheriff, R. E., and L. P. Geldart, 1982, Exploration seismology, Data-processing and interpretation:
Cambridge University Press.
© 2017 SEG Page 5217

High-frequency MASW method and its applications
Zhiqu Lu* National Center for Physical Acoustics, University of Mississippi; Glenn V. Wilson, USDA-ARS National
Sedimentation Laboratory, Watershed Physical Processes Research Unit.
Summary: impossible to identify the fundamental mode of Rayleigh

waves from an overtone image. Thirdly, sensor spacing
A high-frequency MASW method has been recently may become larger than half the wavelength at certain
developed to explore shallow soil in the vadose zone for frequencies, leading to spatial aliases at higher frequencies.
applications in the areas of agriculture, environment, and
military. Several practical techniques were applied to To illustrate this, a typical dispersion curve is plotted in
enhance the HF-MASW method, including (1) the self- Fig. 1(a), along with one wavelength versus frequency. It is
adaptive MASW method using a variable sensor spacing clear that the phase velocity and wavelength decrease
configuration, (2) the phase-only signal processing, and (3) exponentially with frequency. In Fig. 1(b) two curves are
a nonlinear acoustic technique. Using this enhanced HF- plotted: 2.2 wavelengths (solid line) and 0.2 wavelengths
MASW method, the spatial and temporal variations of the (dashed line) versus frequency, respectively. These two
S-wave velocity soil profiles from a few centimeters to a curves are later defined as far offset and near offset for this
few meters were measured. Several case studies were study. Also plotted in Fig. 1(b) is an example of an equal
presented, including (1) soil profiling in the vadose zone, sensor spacing configuration (horizontal lines) with 100
(2) weather and seasonal effects, and (3) fragipan layer sensors and 0.12 cm spacing. With such a sensor
imaging, respectively. configuration, the low frequency part is well sampled
spatially, whereas certain high frequency parts may be
Introduction: under-sampled, failing to meet the Nyquist’s sampling
theorem and introducing spatial aliases at higher
A high-frequency MASW (HF-MASW) method has been frequencies.
recently developed to explore very shallow soil properties
in the vadose zone for applications in the areas of farmland
(a)
management, environmental studies, and battlefield
trafficability evaluations. When the HF-MASW method
was conducted and the collected time traces were processed
in a traditional MASW manner, several technical problems
emerged. First, existing rules for selecting the near offset
and spreadlength cannot satisfy the requirements of planar
dominant Rayleigh waves of the fundamental mode for all
frequencies of interest. These rules were mostly concerned
with the lowest frequency component of surface waves,
rather than all frequency components. At high frequencies,
seismic waves may propagate in far field distance, leading (b)
to predominant higher modes over the fundamental mode
of Rayleigh waves. Secondly, the attenuation coefficients
of Rayleigh waves measured in the HF-MASW test have
much higher values and dynamic range than those of the
conventional MASW methods. This frequency-dependent
attenuation nature prevents all frequency components of
Rayleigh waves from propagating the same distance. On
the other hand, since higher modes of Rayleigh waves have Figure 1: (a) A typical dispersion curve (solid line) and one
longer wavelengths than those of the fundamental mode of wavelength (dashed line) versus frequency, (b) the lengths of 2.2
Rayleigh waves, they penetrate deeper into soil profiles wavelengths (solid line) and of 0.2 wavelengths (dashed line)
(Xia, et al., 2003) and consequently are less attenuated than versus frequency, and an example of an equal sensor spacing
the fundamental mode. At certain high frequencies and configuration (horizontal lines) with 100 sensors and 0.12 cm
distances, the energy of Rayleigh waves of the fundamental spacing to cover a spreadlength of 12 meters.
mode will not dominate over those of the higher modes of
Rayleigh waves, making it difficult and sometimes To overcome the above problems, a self-adaptive method
with a variable sensor spacing configuration was developed
© 2017 SEG Page 5218

High-frequency MASW method
to determine high frequency dispersion trends (Lu, 2014b; The term of (nfar – nnear+1) in Eq. (3) is used for
Lu, 2015; Lu, 2017), which is a minor modification of the normalization.
method proposed by Park and Rydén (2007) and Park

(2011), which they named as the selective offset method. Furthermore, in order to avoid spatial aliases and minimize
the number of sensors, a variable sensor spacing
Enhanced high-frequency MASW method: configuration is implemented which usually features two
sensor arrays: a fixed spacing sensor array and a
The self-adaptive HF-MASW method with a variable progressively increased sensor spacing array.
sensor spacing configuration
The self-adaptive HF-MASW method with a variable Due to the self-adaptive method, the summation in Eq. (3)
sensor spacing configuration has been described in detail in is made over a short spreadlength with small spaces at a
the literature (Lu, 2015). For a fixed-offset and full high frequency and over a long spreadlength with large
spreadlength MASW method, an overtone image is spaces at a low frequency.
obtained by the 2-D wavefield transformation method
proposed by Park et al. (1998) and its summation form is The phase-only signal processing
expressed as follows: For the phase-only signal processing, the magnitude term
e R ( f ,x ) , (1) Rj(f, xj) in Eq. (1) and Eq. (3) is simply set to 1 and only
N i
E ( f , c)  j 1
j
j j
the phase term is taken into the waveform transformation.
where f and c are frequency and phase velocity, E(f, c) the Therefore the influence of the frequency and distance on
energy at the coordinates of (f, c) in phase velocity ~ the magnitude is eliminated as long as the phase spectrum
frequency space, Rj(f,xj) the magnitude term of the fast- is reliably measured. Because of this phase-only approach,
Fourier-transform of the jth time trace at offset xj, phase the constants of 0.2 and 2 are set for the near-offset and
∅j=2πfxj/c, N the number of the time traces, and i denotes spreadlength in Eq. (2), avoiding the requirement of
the imaginary unit. In this formulation, summation is made adjusting these coefficients to mitigate the concern of
over the entire set of time traces and the Fourier transform magnitude frequency-dependency (Lu, 2015).
Rj(f,xj) is usually normalized to unit amplitude for each
trace. The self-adaptive survey geometry along with the phase-
only processing technique is somewhat similar to the
For a self-adaptive or selective offset MASW method, the spectral analysis of surface wave (SASW) method
near offset xnear (the distance between a seismic source and (Nazarian and Stokoe, 1984) in data acquisition (employing
the first selected sensor), spreadlength L (the length variable spreadlength at different frequencies) and signal
covered by the selected subset of sensors), and the far processing (considering only phase). However, the current
offset xfar (the distance between a seismic source and the HF-MASW is superior to the SASW method in sensor
final selected sensor) are determined by the wavelength λ at deployment and mode identification. Besides, in the phase-
each frequency: only MASW method, no unwrapping of the phase is
xnear  0.2 ( f ) required, whereas the SASW is vulnerable to noise in
, (2)
L  2 ( f ) unwrapping the phase.
x far  xnear  L
At each frequency, the near offset and spreadlength are Nonlinear phenomenon
unique at each frequency and self-adaptive to the In a routine HF-MASW test, the soil surface is excited by
corresponding wavelength. Eq. (2) uses two constants of an electrodynamic shaker operating in a frequency
0.2 and 2 which were adjustable coefficients in the previous sweeping mode. It is found that the measured dispersion
equations (Park and Rydén, 2007; Park, 2011; Lu, 2014b; curves are well resolved at significantly higher frequencies
Lu, 2015). The reason of using these constants will be than the source frequency bandwidth. It is speculated that
explained later. due to nonlinear effects up to the 3rd harmonics could be
generated. There are two possible nonlinear mechanisms:
The 2D wavefield transformation is performed by summing (1) soil nonlinearity-a so-called mesoscopic hysteretic
over the time traces that satisfy Eq. (2) and can be elasticity (Lu. 2005; Lu, 2007; Guyer and Johnson, 2009)
expressed as (Park, 2011): and (2) contact nonlinearity generated between the shaker
and soil surface. These nonlinear effects extend the

n fa r i
j nn ea r
e j Rj ( f , x j ) , (3) detectable frequency range of the dispersion curves and
E ( f , c) 
( n far  nnear  1) practically benefit data acquisition process by reducing
where nnear and nfar are indices of the time traces that match frequency bandwidths with gapped frequency bands. Doing
closely the near offsets xnear and far offset xfar, respectively. so would increase seismic energy into the ground by
maintaining the same chirp duration.
© 2017 SEG Page 5219

Experimental Setup: method. The results from both the HF-MASW method and
soil characterizations were in good agreement, validating
The high-frequency MASW method consists of an that the current HF-MASW method can be used to in situ
electrodynamic shaker (Vibration Test System, Model VG- non-invasively explore the soil profile in the vadose zone.
100-6) and vibration sensors: either a moving laser Doppler
vibrometer (LDV, Polytec PI, Inc., Model PDV 100,
frequency range: DC-22 kHz)) and an accelerometer (PCB
Piezotronics, Model 352B, 2 to 10000 Hz) or a geophone
array (40 geophones, 40 Hz, GS-20DM, Geospace
Technologies). The shaker operates in a frequency-
sweeping mode to generate a chirp signal with three or four
overlapped or gapped frequency bands, representing low-
frequency (LF), middle frequency (MF), high frequency
(HF), and extra high frequency (XF) bands, respectively.
For example, typical four gapped frequency bands are 20-
60 Hz, 80-300 Hz, 500-800 Hz, and 1000-1600 Hz,
respectively. The dispersive images obtained from these
frequency bands are combined to form a dispersive image
that covers the whole frequency range of interest. A typical
sensor spacing configuration is described as follows: the
first 120 sensor spaces are set to be an equal space of 0.5
cm and the next 40 sensor spaces start with 1 cm spacing
followed by a 1 cm incremental spacing for each
subsequent sensor. The frequency bands and sensor spacing
configurations can be adjusted according to the specific
task of the each test.
The applications of the HF-MASW method
Using this enhanced MASW method, the spatial and

temporal variations of soil properties from a few
centimeters to a few meters can be measured. Several case
studies are presented as described below.
Soil profiling in the vadose zone

The HF-MASW tests were conducted at the North Figure 2: The soil profiles for (a) dry sealed/crusted, (b) moist
Mississippi Agriculture and Forestry Experiment Station at non-tilled, and (c) moist tilled loess soils, respectively. Note: in
Holly Springs, Mississippi on a Providence silt loam (fine- Fig.2(a), the solid line is the profile inverted by SurfSeis3 (Version
silty, mixed, active, thermic Oxyaquic Fragiudalfs) under 3.0.6.4, Kansas Geological Survey), whereas the round dots and
three soil conditions, i.e. a naturally formed dry circled dots represent the profiles inverted by a simple
approximation method. In Fig. 2(b)-(c), the soil profiles are
sealed/crusted, a moist non-tilled, and a moist tilled soil
inverted by SurfSeis3.
condition (Z. Lu and G.V. Wilson, 2017a, personal
communication). The soil profiles in terms of the shear
wave velocity were measured up to 1.5 meters deep below Weather and seasonal effects study
the surface, as shown in Fig. 2. On the dry sealed/crusted A long-term survey was conducted to study weather and
soil, three distinctive zones were identified, representing a seasonal effects on subsurface soil (Lu, 2014a). The test
high velocity surface sealed/crusted thin layer, a moist and was conducted on a campus site of the University of
soft middle zone, and a progressively increased velocity Mississippi where the original soils of a thickness of about
region below mainly caused by increased overburden 10 feet were excavated and refilled with washed sands.
pressure with depth. On moist non-tilled and tilled soils, This site can be regarded as a homogeneous medium
the high velocity top layer disappeared and exhibited a without the presence of distinctive layers, a favorable
generally increasing tendency with depth with subtle condition for the task. During a one-year long survey, the
differences among non-tilled and tilled conditions. Soil temporal variations of soil temperature, water content,
core samples, shear vane and penetration tests were water potential, and P-wave velocity were monitored by
compared with the observations from the HF-MASW buried sensors, as plotted in Figs. 3 (a)-(d). The temporal
© 2017 SEG Page 5220

variations of soil profiles measured by the HF-MASW was conducted on the North Mississippi Experiment
method were shown in Figs. 3(e)-(f), which reflected and Station at Holly Springs, MS (Lu and Wilson, 2017b) using
matched the temporal changes of soil conditions due to both a two-dimensional HF-MASW test and invasive soil
weather and seasonal effects, especially for those of water profile characterization such as penetration tests. Fig. 4
potential. It is remarkable that the most significant changes showed the spatial distributions of the S-wave velocity of
in the soil profile images occurred during the dry summer the vertical cross-section in the testing site. From the
as shown in Fig. 3(f). The dramatic changes in the S-wave contrast of the S-wave velocity image, the presence, depth,
velocity in magnitude, extent, and depth happened not only and extent of a fragipan layer were determined, manifested
in the top zone but in the deeper zone as well, indicating an as a high velocity layer. The results from both the HF-
expanded influence of dry soil condition in the summer that MASW and penetration tests were compared and they were
penetrated into deeper soils. These observations can be in good agreement. It was also found that the addition of
understood and interpreted with the concept of the effective the first higher mode of Rayleigh waves in the inversion
stress, governed by soil suction stress/water potential for process significantly improved the accuracy and resolution
top unsaturated soils and by overburden pressure for deeper of the S-wave velocity image, especially in identifying
layers of soils (Lu, et al., 2004; Lu and Sabatier, 2009). structural layering information and spatial heterogeneity.
Figure 4: The S-wave velocity cross-setion image of the test site,

where the high velocity layer detemines the presence, depth, and
extent of a fragipan layer
Conclusions
A HF-MASW method with the enhanced techniques such

as the self-adaptive method with a variable sensor spacing
configuration, the phase-only processing, and nonlinear
technique was described. Using this enhanced MASW
method, the spatial and temporal variations of soil
properties from a few centimeters to a few meters below
the surface were measured. It has been demonstrated that
the HF-MASW method can be applied for measuring soil
profiles in the vadose zone, monitoring subsurface soil
variations due to weather and seasonal effects, and
Figure 3: The temporal varaitions of soil conditions and soil detecting and imaging a fragipan layer. Future work may
profiles. (a) soil temperature, (b) water content, (c) water potential, include agricultural compaction and monitoring
(d) P-wave velocity, (e) soil profiles in the spring (f) soil profiles instantaneous responses of soil profile during rainfall or
in the summer. For the sake of brevity, the soil profiles in the irrigation events.
winter were not displayed in Fig. 3(e)-(f).
Acknowledgements
Fragipan detection
A fragipan is a naturally occurring soil layer with very low This work was supported by the U.S. Department of
organic matter, high bulk density and mechanical strength, Agriculture under Non-Assistance Cooperative Agreement
and hard consistence when dry, but brittle when moist. It is 58-6060-6-009.
restrictive to root and water penetration. Fragipans play a
critical role in hydrologic behavior, erosion, and land use.
In order to detect and image a fragipan layer, a field test
© 2017 SEG Page 5221

EDITED REFERENCES
REFERENCES
Guyer, R. A., and P. A. Johnson, 2009, Nonlinear mesoscopic elasticity: The complex behavior of
granular media: WILEY-VCH Verlag GmbH & Co. KGaA.
Lu, Z., 2005, Role of hysteresis in propagating acoustic wave in porous materials: Geophysical Research
Letters, 32, L14302, http://doi.org/10.1121/1.4784955.
Lu, Z., 2007, The phase shift method for studying nonlinear acoustics in soils: Acta Acustica united with
Acustica, 93, 542–554.
Lu, Z., 2014a, Feasibility of using a seismic surface wave method to study seasonal and weather effects
on shallow surface soils: Journal of Environmental & Engineering Geophysics, 19, 71–85,
http://doi.org/10.2113/JEEG19.2.71.
Lu, Z., 2014b, Self-adaptive method for high-frequency dispersion curve determination: Proceedings of
the Symposium on the Application of Geophysics to Engineering and Environmental Problems
(SAGEEP 2014).
Lu, Z., 2015, Self-adaptive method for high frequency multi-channel analysis of surface wave method:
Journal of Applied Geophysics, 121, 128–139, http://doi.org/10.1016/j.jappgeo.2015.08.003.
Lu, Z., 2017, Practical techniques for enhancing the high-frequency MASW method: Journal of
Environmental and Engineering Geophysics, 22, 197–202.
Lu, Z., C. J. Hickey, and J. M. Sabatier, 2004, Effects of compaction on the acoustic velocity in soils: Soil
Science Society of America Journal, 68, 7–16, http://doi.org/10.2136/sssaj2004.7000.
Lu, Z., and J. M. Sabatier, 2009, Effects of soil water potential and moisture content on sound speed: Soil
Science Society of America Journal, 73, 1614–1625, http://doi.org/10.2136/sssaj2008.0073.
Lu, Z., G. V. Wilson, and C. J. Hickey, 2014, Imaging a soil fragipan using a high-frequency MASW
method: Proceedings of the Symposium on the Application of Geophysics to Engineering and
Environmental Problems (SAGEEP 2014).
Lu, Z., and G. V. Wilson, 2017b, Imaging a soil fragipan using a high-frequency multi-channel analysis
of surface wave method: Journal of Applied Geophysics, 143, 1–8,
https://doi.org/10.1016/j.jappgeo.2017.05.011.
Nazarian, S. and K. H. Stokoe II., 1984, In-situ shear wave velocities from spectral analysis of surface
waves: Proceedings of the 8th Conference on Earthquake Engineering, 31–38.
Park, C. B., 2011, Imaging dispersion of MASW data-full vs. selective offset scheme: Journal of
Environmental and Engineering Geophysics, 16, 13–23, http://doi.org/10.2113/JEEG16.1.13.
Park, C. B., and N. Rydén, 2007, Offset selective dispersion imaging: Proceedings of 2007 SAGEEP,
910–915.
Xia, J., R. D. Miller, C. B. Park, and G. Tian, 2003, Inversion of high frequency surface waves with
fundamental and higher modes: Journal of Applied Geophysics, 52, 45–57,
https://doi.org/10.1016/S0926-9851(02)00239-2.
© 2017 SEG Page 5222

Timelapse Surface Wave Monitoring of Permafrost Thaw Using Distributed Acoustic Sensing
and a Permanent Automated Seismic Source
J. Ajo-Franklin*1, S. Dou1, N. Lindsey1,2, T.M. Daley1, B. Freifeld1, E.R. Martin4, M. Robertson1, C. Ulrich1, T.
Wood1, I. Eckblaw1, and A. Wagner3,
1
Lawrence Berkeley National Laboratory, 2 University of California, Berkeley, 3 Cold Regions Research &
Engineering Laboratory, 4 Stanford University
during thaw are available. To test the potential applicability
Summary of timelapse surface wave measurements for tracking the
We present preliminary results from an intermediate scale thaw process proceeding subsidence, we utilized a fully
field experiment exploring the seismic response of dynamic automated seismic source (e.g. Freifeld et.al. 2016, Ikeda
permafrost thaw generated by active heating. The focus of et.al. 2017) recorded on trenched fiber optic cables using
our project was to evaluate the utility of surface wave distributed acoustic sensing (DAS) (e.g. Daley et.al. 2013).
monitoring to detect precursors to thaw-induced This system was installed over a zone of well-characterized
subsidence, a common geotechnical hazard in polar permafrost. A 10 x 13 m section of permafrost was thawed
regions. In this study, we present results from timelapse over a period of 2 months to evaluate the seismic response
surface wave measurements conducted over the duration of during and after subsidence.
the thaw experiment. The unique aspect of the experiment
Field Site & Installation
was the combination of a semi-permanent surface orbital
The experiment was conducted at the Fairbanks Permafrost
vibrator (SOV) source and distributed acoustic sensing to
Experiment Station (Farmer’s Loop Site), near Fairbanks,
measure variations in surface wave propagation. The SOV,
AK. The site is in a zone of discontinuous permafrost
energized for 22 sweeps every night, was deployed for
dominated by silty soils and administrated by the Army
approximately 2 months, collecting 60 daily surveys. Large
Corp of Engineers, Cold Regions Research and
temporal variations in surface wave velocity, as well as
Engineering Laboratory. The permafrost table topography
spectral characteristics were observed. After examination
varies between ~1.5 m and 4.5 m depth across the site,
of precipitation and soil moisture data, such changes were
dipping to the west due to removal of vegetation. The
convincingly linked to rainfall events. A cross-equalization
heaters and geophysical systems were installed over the
approach was developed to assist in removing this effect;
summer of 2016. The controlled thaw system consisted of
after processing, a decreasing trend in shear wave velocity
an array of 121 resistive borehole heaters, installed close to
appears to remain, potentially a seismic signature of the
the permafrost table at a depth of between 3 and 4 m. The
controlled permafrost thaw process.
heaters were installed in a grid, with approximately 1 m
between heaters, covering a 10 x 13 m surface footprint.
Introduction
The heaters were connected into 10 sub-arrays and
In areas of ice-rich permafrost, thaw induced by increasing
collectively drew a constant 3600 W of power during the
atmospheric temperatures can lead to sudden surface
thaw experiment. Figure 1 shows a schematic map of the
subsidence, a process which presents a geotechnical hazard
site with the heater array visible as red dots. In addition to
to critical Arctic infrastructure. Unfortunately, few cost
the heaters, a wide range of secondary systems were
effective approaches exist to detect zones of permafrost that
installed including borehole temperature arrays (thermistors
are on the brink of mechanical failure, particularly in
and thermocouples), ERT electrode arrays, a geophone
locations of discontinuous permafrost where the permafrost
array, broadband seismometers, and integrated fiber-optic
table is already in flux. Seismic properties are one
cables in trenches and boreholes.
geophysical parameter sensitive to pre-thaw changes in
A surface orbital vibrator (SOV) (see Freifeld et.al.
mechanical properties; significant variations in shear wave
2016) was used as a repeatable seismic source for the
velocity (Vs) occur well before 0 C (Li, 2009) due to the
MASW monitoring component of the experiment. SOVs
variable freezing point of water in finer grained sediments
consist of a rotating eccentric mass spun to high velocity;
(Romanovsky and Osterkamp, 2000). With the availability
the resulting sweeps have been shown to be repeatable in
of increasingly inexpensive seismic recording approaches
past studies. In this case, the SOV was a small industrial
(e.g. nodal stations and distributed acoustic sensing) as well
model with a maximum force of 2 tons. The SOV was
as techniques based on utilization of ambient infrastructure
bolted to a 1x1x1 m concrete foundation; the system was
noise as a source, the use of surface wave monitoring for
fully automated to run in an unattended manner controlled
detecting zones of permafrost alteration seems within
from a remote PC. Figure 2 shows the installation of the
reach.
SOV (a), final configuration (b), and an interior photograph
While past MASW studies have demonstrated low
of the eccentric mass (c), and a schematic showing the
Vs values in zones of permafrost subsidence (Hayashi et.al.
clockwise (CW) and counter-clockwise (CCW) spin
2011), no timelapse measurements tracking Vs anomalies
orientation. A 3C monitor geophone was installed 1 meter
© 2017 SEG Page 5223

Timelapse Surface Wave Monitoring of Permafrost using DAS
beyond the SOV pad to provide a reference signal observed at the surface by total station surveys and LiDAR
recording. (not shown).
Figure 2 : SOV installation: (a) SOV foundation construction, (b)

final SOV geometry and control unit, (c) mass configuration, and
(d) rotation schematic.
MASW surveys were acquired each night using the SOV
source between 11pm and 2am to allow ambient noise
recording during peak traffic hours. Each nightly survey
consisted of 22 clockwise and counterclockwise sweeps in
Figure 1 : DAS/SOV installation at the Farmer's Loop Rd addition to several “warm-up” sweeps to improve
Permafrost Station. Note the primary DAS lines (A-D) and the repeatability. Each sweep had a total duration of 180s
location of the SOV source (red star). including a 30s acceleration and a 150s spin-down. The top
Controlled source MASW data as well as ambient noise frequency of each sweep was 90 Hz with a 60 Hz median
data were acquired using a 2D array of trenched fiber optic frequency. Figure 4, panels (a) and (b) show a single sweep
cables. The cable selected was a polyeurethane-jacketed as recorded in the time-domain by the pilot geophone as
tight-buffered tactical cable manufactured by OCC which well as a time-frequency plot showing the sweep stages.
included 6 single-mode (SM) and 4 multi-mode (MM) DAS data was recorded continuously for the entire
fibers allowing simultaneous acquisition of distributed duration of the experiment using a Silixa iDAS
acoustic (DAS), temperature (DTS), and strain (DSS) data. interrogator. Data was sampled at 1 kHz and spatially
The shallow (30 cm) trenches were back-filled with native discretized at 1m; the iDAS unit had a gauge length of 10
soil. As shown in figure 1, 5 short lines (1-5), 4 longer (180 m. DAS is most sensitive to extensional strain parallel to
m) lines (A-D), and a 500 m off-pad line were emplaced for the fibe; the spin plane of the SOV is oriented parallel to
a total 7 km of fiber. The fiber grid was oriented so that the lines A-D to allow effective recording of S and surface
shorter lines were orthogonal to the dominant seismic noise waves sampling the thaw region.
source (Farmer’s Loop Rd). Figure 3 shows two trench Data Processing
sections before backfill (a,b) as well as spooling operations
during installation (c). Due to relative clock drift between the SOV control system
and the iDAS, the first processing step involved a time-
Data Acquisition frequency domain alignment between the pilot geophone
The MASW component of the project was conducted and a near-source DAS channel. After time corrections, a
between August 4th and October 2nd, 2016, concurrent with time-varying filter matched to the sweep was applied to
the subsurface heating experiment. During the course or reject secondary noise, followed by deconvolution using
heating, soil temperatures in the 3-4 m range increased the Y (in-line) pilot geophone component.
from near 0 C to ~25 C, permafrost table depth was
deepened by ~1-1.5 m, and up to 10 cm of subsidence was
© 2017 SEG Page 5224

packet and a faster S refraction off of the top of the

permafrost table are visible. The frequency content of the
survey was less than expected at intermediate to far offsets

with most energy in the 10-30 Hz band with the exception
of higher frequencies in the refracted arrivals.
Figure 3 : Trenching and fiber installation. Panels (a) and (b) show
trench character in two sub-regions of the site. Panel (c) depicts
fiber installation.
Figure 5 : Precipitation related changes in source behavior. Top

panel shows spectral variations in the source as recorded by the
monitor geophone while the middle and bottom plots show soil
moisture and rainfall data recorded at the site.
Rainfall and Soil Moisture Observations

Early inspection of field records suggested large temporal
variability in the source signature over the day-to-week
timescale as recorded by the pilot geophone, despite good
sweep-to-sweep repeatability. Figure 5 (top) shows spectral
variations in the pilot geophone records over the duration
of the study. The dominant feature was the appearance of a
variable spectral notch as well cyclic velocity reductions
observed in the surface wave arrivals. These seismic
variations were traced back to large changes in soil
moisture after rainfall events (Figure 5, middle & bottom)
and subsequent drying periods. The flat terrain and poor
drainage of the site likely contributed to these effects which
unfortunately obscure longer-term signatures potentially
related to the induced thaw process.
Figure 4 : Sample SOV sweeps and deconvoled gather. Panels (a)
and (b) show the time domain sweep and time-frequency Cross-equalization and Traveltime Analysis
representation respectively as recorded by the monitor phone. To mitigate the impact of these moisture-driven processes,
Panels (c) shows a deconvolved SOV gather with frequency we utilized a cross-equalization approach based on Wiener
content displayed in panel (d). filters. Between the four 180 meter- long surface DAS lines
After deconvolution, CW and CCW sweeps were that are used for monitoring, line A is entirely outside the
separately stacked into nightly epochs (22 sweeps). Figure heater array and is expected to be least affected by the
4, panel C shows the result of one nightly stack in the time heating. Here we assume: (1) line A is unaffected by the
domain for line C. As can be seen, a slow surface wave heated area; (2) precipitation-induced data variations are
© 2017 SEG Page 5225

identical among these four lines (A, B, C, and D). We then fellowship grant number DE-FG02-97ER25308. N.
constructed Wiener filters using data from line A and Lindsey was supported by an NSF Graduate Fellowship.
applied the filters to lines B-D. We would like to thank Stan Wullschleger for his initial
The cross-equalization procedure is applied in the soil heating prototype as well as CRREL personnel for
time domain. The input data are shot gathers in which the critical assistance during installation.
SOV source signatures are already removed via
deconvolution. We use line A data from the first sweep of
the first day as the template. For the remainder of the pre-
stack sweeps, we construct Wiener-filter coefficients that
are specific to each DAS channel; this procedure is a time-
domain least-squares implementation of spectral shaping.
After applying the resulting Wiener filters to all four lines,
post cross-equalization shot gathers are stacked on a daily
basis, and the resulting post-stack shot gathers are used in
subsequent time-lapse analyses.
Figure 6 shows the surface wave window for
single DAS channels on the different lines (A-D) at the
north end of the thaw zone before (a) and after (b) the
cross-equalization procedure. As can be seen, cross-
equalization removes the majority of the rainfall-related
changes in arrival time. Line A (the reference line) is of
course invariant after cross-equalization.
Figure 7 compares time-lapse changes in surface
wave velocity, dv/v0 with and without cross-equalization,
where dv/v0 is the relative phase perturbation estimated
using cross-correlation of the dominant Rayleigh-wave
window (first 1.5 cycles). Before applying cross-
equalization (Fig. 7 a), dv/v0 is dominated by the influences
of precipitation, which are manifested by the anti-
correlations between dv/v0 and soil-water content. After
cross-equalization (Fig. 7 b), systematic trends of velocity
reduction of up to 6%, likely due to permafrost thaw, are
visible for line C and D. The limitation of this analysis is
the strong assumption that precipitation-induced seismic
variations are identical among all four lines and that line A Figure 6 : Timelapse surface wave traces for fixed channels at the
is not influenced by the heating plot. north edge of the heated zone before (A) and after (B) the cross-
equalization procedure.
Conclusions
In this study we have demonstrated the utility of distributed
acoustic sensing, integrated with an automated seismic
source, for collecting timelapse surface wave monitoring
data in a permafrost environment. The fine temporal
sampling of the survey allowed identification of
precipitation-induced velocity variations that dominated
response at our experiment location. After removal of soil
moisture effects using a cross-equalization procedure, a
long-term shear wave velocity reduction is apparent. We
currently believe this trend is a seismic signature of the
induced thaw process; on-going analysis of the SOV shear
refraction data, ambient noise measurements, and ERT
should assist in validating this hypothesis.
Acknowledgements
This study was funded by the Strategic Environmental Figure 7 : Surface wave traveltime estimates at the north end of the
Research and Development Program (US DoD), grant RC- heating plot for each profile before (a) and after (b) the cross-
equalization procedure.
2437. Eileen Martin was supported by DOE CSGF
© 2017 SEG Page 5226

EDITED REFERENCES
REFERENCES
Daley, T. M., B. M. Freifeld, J. Ajo-Franklin, S. Dou, R. Pevzner, R. Shulakova, S. Kashikar, D. E.
Miller, J. Goetz, J. Henninges, and S. Lueth, 2013, Field testing of fiber-optic distributed sensing
(DAS) for subsurface seismic monitoring: The Leading Edge, 32, 699–706,
Dou, S., and J. B. Ajo-Franklin, 2014, Full-wavefield inversion of surface waves for mapping embedded
low-velocity zones in permafrost: Geophysics, 79, no. 6, EN107–EN124,
Freifeld, B. M., R. Pevzner, S. Dou, J. Correa, T. M. Daley, M. Robertson, K. Tertyshnikov, T. Wood, J.
Ajo-Franklin, M. Urosevic, and B. Gurevich, 2016, The CO2CRC Otway Project deployment of a
distributed acoustic sensing network coupled with permanent rotary sources: 78th Annual
International Conference and Exhibition, EAGE, Extended Abstracts,
http://dx.doi.org/10.3997/2214-4609.201600577.
Hayashi, K., J. Tunnicliffe, A. Becker, P. Finlay, L. Schrott, and C. Burn, 2011, Integrated multi-
instrument surveys of permafrost conditions under a runway, Inuvik, NWT, Canada: Presented at
the Society of Exploration Geophysics, Annual Meeting (Permafrost Workshop), San Antonio,
TX.
Ikeda, T., T. Tsuji, M. Takanashi, I. Kurosawa, M. Nakatsukasa, A. Kato, K. Worth, D. White, and B.
Roberts, 2017, Temporal variation of the shallow subsurface at the Aquistore CO2 storage site
associated with environmental influences using a continuous and controlled seismic source:
Journal of Geophysical Research: Solid Earth, 122, 2859–2872,
http://dx.doi.org/10.1002/2016JB013691.
Li, H., 2009, Experimental and numerical study of sonic wave propagation in freezing sand and silt: Ph.D.
thesis, University of Alaska, Fairbanks.
Romanovsky, V. E., and T. E. Osterkamp, 2000, Effects of unfrozen water on heat and mass transport
processes in the active layer and permafrost: Permafrost and Periglacial Processes, 11, 219–239,
http://dx.doi.org/10.1002/(ISSN)1099-1530.
© 2017 SEG Page 5227

MASW Control of Grouting
Alessandro Cirone* and Roger Rodrigues, Engegraut Ltda, and Choon Park, Park Seismic LLC
Summary
Because of the importance and extension of the project, all
Geophysical surveys using the MASW method were worked areas were submitted to a rigid quality control
conducted over a site in Pontal Oceânico neighborhood, program. One of them also received a geophysical
Rio de Janeiro, Brazil, during the construction of the Rio investigation with MASW surveys. Running geophysical
Olympic Media Village. Soft clay soils were found in this tests over treated and untreated areas represented an
region, requiring to be improved by the use of a innovative attempt to evaluate ground modification efficiency in terms
grouting technique called CPR Grouting. Since the primary of increase in shear wave velocity. In addition, the authors
objective of the ground improvement was to provide performed the tests with the aim of setting the standards for
strength and limit settlements, quality control appeared soil improving acceptance for future applications.
more rational in terms of minimum modulus rather than of Compared to the conventional in situ tests, that can
minimum penetration resistance. The MASW method measure only locally the soil properties, the shear wave
evaluates subsurface shear-wave velocity (Vs) distribution velocity approach represented the only viable method in a
by analyzing seismic surface waves that always maintain a situation in which a large volume of soil needed to be
strong signal-to-noise (SN) ratio. The shear-wave velocity investigated with a relatively dense spatial sampling
(Vs) is a direct indicator of material's stiffness. Therefore, interval.
the purpose of the MASW surveys was to produce shear-
wave velocity (Vs) profiles (1-D and 2-D) of natural and Multichannel Analysis of Surface Waves (MASW)
improved ground in order to evaluate the grouting
efficiency by comparing subsurface velocity (Vs) The MASW method (Park et al., 1999) is a seismic surface-
distributions. Processing results of 2-D velocity (Vs) cross wave technique developed specifically for near-surface
sections obtained from natural (untreated) and nearby applications at depths usually shallower than a few tens of
improved (treated) grounds show velocity (Vs) increase by, meters (e.g., ≤ 30 m). Since its first introduction in the late
up to, 50% for the depths of grouting observed throughout 1990s, use of the technique has rapidly increased for two
the entire lateral survey distance. This case study strongly reasons: (1) it provides the shear-wave velocity (Vs) of
indicates the seismic (MASW) method can provide a robust ground materials, which is one of the most important
tool to evaluate and control the effectiveness of the geotechnical parameters in civil engineering, and (2) it is
grouting technique. the easiest seismic survey method because of the high
signal-to-noise (SN) ratio always ensured. MASW can
Introduction produce Vs profiles in 1-D (depth) or 2-D (depth and
surface) modes.
Pontal Oceânico is an entire new neighborhood located in
the Recreio district, covering an area of 600 000 m², on the MASW provides Vs information of ground materials by
west side of Rio de Janeiro. It was planned to be the processing Rayleigh-type surface waves that are dispersive
Olympic Media Village during the Rio 2016 Olympic when travelling through a layered media (different
Games and then transformed into a residential area. frequencies travel at different speeds). This dispersion
Buildings needed infrastructure works including property is determined from a material's shear-wave
excavations, embankments for access roads and the velocity (Vs) (by more than 95%), P-wave velocity (Vp) (≤
building of a canal system for managing surface water 3%), and density (ρ) (≤ 2%). By analyzing dispersion
drainage. properties, we can determine Vs fairly accurately by
assuming some realistic values for Vp and ρ.
A series of field and laboratory tests were carried out for
geotechnical site characterization, identifying low bearing Stiffness of a material is defined as a measure of resistance
capacity and highly compressible soft clay deposits in the to deformation (Sheriff, 2002) and ultimately related to
area. Earthwork structures built on the weak subsoil material's elastic moduli that describe the material's
required a foundation solution against potential failures and behavior under stress. Among the three primary types of
large consolidation settlements. Design engineers elected to moduli – Young's (E), shear (G), and bulk (K) – according
use soil modification with grouting because of a potential to the theory of elasticity (Sheriff and Geldart, 1982), the
reduction in overall construction time. They opted for a shear modulus, G, is related to the shear wave velocity of
brazilian grouting method called CPR Grouting, the soil, Vs, through the basic expression:
specifically developed to modify the geotechnical
properties of soft clays.
© 2017 SEG Page 5228

G  ρVs2 (1) applications can be found in Riccio et al. (2013) and Cirone
(2016). Here, for clarity, construction stages are briefly
This means that a reliable determination of G can be resumed.

obtained by measuring Vs. This is why seismic shear-wave
velocity (Vs), the final product from an MASW survey, is Firstly, prefabricated vertical drains (PVDs) are installed
often used as a direct indicator of a material's stiffness. throughout the soft soil deposit in a square or triangular
pattern. Then, compact-shape grout bulbs are expanded by
Site Conditions means of sequential subsurface injections. The grout used
is a low-slump mortar, a sand–silt–cement mixture with
The site is located in the lowland west side of Rio de high internal viscosity to prevent hydraulic fracturing (Au
Janeiro, in the proximity of the ocean. Besides the natural et al., 2003). The soil surrounding the bulbs is displaced
variation in soil condition, the site had a definite overall and at the same time compressed. The consolidation
layer sequence as shown in Figure 1. Site investigation process that follows the grouting stage provides an increase
identified alluvial and marine deposits, consisting of an in effective stress. As a result, grouted soil shows high
upper soft clay layer underlain by sand. In some areas, a horizontal confining stress, and can be assumed as a
superficial old earth fill was encountered. Residual soil was composite material, in which the grout bulbs represent the
found at greater depths, in general below 25 m. In the reinforcement. Soil strength, stiffness and bearing capacity
region, soft soils typically have high organic matter are enhanced.
content, with presence of peat, root, shells and fine layers
of sand and silt. They are characterized by high water
Concrete Grout injection
content, high compressibility, low strength, and poor
truck equipment
permeability. The geotechnical properties of Rio de Janeiro
soft clays have been widely investigated during the past
two decades by many researchers (Almeida, 1998;
Sandroni, 2006; Baroni, 2010). The upper soft layer of soil
is usually extremely soft, with NSPT blow count value less
than 1 and cone penetration resistance below 200 kPa
(Sandroni and Consoli, 2010). Grout pump
Figure 2. Grouting work in progress at Pontal Oceânico.
Quality control is performed according to a well-defined

procedure, consisting of slump tests, pressure and injection
rate recordings, excess pore water pressure monitoring, and
in situ tests. Pressuremeter test are analyzed according to
Briaud (1992) to evaluate the increase in soil stiffness after
grouting.
Although the pressuremeter is based on the same principle

as the ground modification method, that is, both consist of a
cavity expansions (see Yu, 2000), the test remains a local
measurement of soil stiffness. As a matter of fact, common
soundings (SPT, CPT), as well as pressuremeter tests,
Figure 1. Generalized subsurface profile. cannot provide a macroscale measurement of soil stiffness.
Results are always highly dependent on test location,
Ground Improving Method because the actual improved ground is a composite
material. In this context, geophysics provided a flexible
The main geotechnical requirement related to the tool for the macroscale assessment of grouting efficiency,
realization of the Pontal Oceânico neighborhood was by relating the increase in shear wave velocity to the gain
improving the upper soft clay deposit, reaching an average in soil stiffness. A similar approach has been reported by
depth of 10 m. This was made by the use of the CPR Park and Miller (2004) for the evaluation of soil stiffness
Grouting, which is a grouting technology specifically before and after mechanical surface compaction, at a site
designed for consolidating soft clays. The photograph in with soil consisting of a mix of gravel and cobbles in a
Figure 2 shows the grouting stage in progress. More sand-and-silt matrix.
information about its geotechnical modeling and
© 2017 SEG Page 5229

MASW Surveys (Untreated and Treated Grounds) The final output of 2-D velocity (Vs) cross section at the
untreated site (Figure 3) shows that the lowest velocity,
The MASW method was used with active sources in order below 100 m/sec, occurs at a depth of about 6 m. On the
to obtain 1-D and 2-D Vs profiles representing subsurface other hand, the 2-D velocity (Vs) cross section at the
variation of soil stiffness. In particular, MAWS was chosen treated (improved) site (Figure 4) shows an overall
for being able to characterize soil profile with inverse soil subsurface velocities of about 150 m/sec for depths down
layers. The overall procedure consisted of field data to about 7 m. There was a thick (~1.5 m) supporting layer
acquisition by using a 24-channel seismograph and 4.5-Hz of harder materials built on the surface of the untreated site
geophones, dispersion curve processing, inversion analysis to facilitate the movement of construction vehicles. This
and comparison of results. layer is noticed by the high velocity (e.g., Vs ≥ 300 m/sec)
on top of the velocity (Vs) cross section in Figure 3.
The surveys were performed at the beginning of December,
2015. The aim of the first field day was to practice the Finally, cross sections are compared with a velocity (Vs)
entire deployment of acquisition gears, and also to collect difference plot. There was about 1.5 m difference in surface
records to produce 1-D Vs profiles. A few different source level between lines 1 and 3, due to the presence of the
offsets were used for each stationary array to increase data supporting layer on top of the untreated site. For an
quality during the dispersion curve processing. The second objective comparison, the depth values in the line 1 cross
day of flied survey used the land streamer of 1.0 m receiver section have been corrected by 1.5 m to account for the
spacing, and performed one 2-D production line at an difference in surface levels between the two sites. Results
“untreated” soil site (line 1). It was decided to reduce the are reported in Figure 5, showing a noticeably velocity
receiver spacing from 1.5 m to 1.0 m, since maximum increase up to 90 m/sec within the improved layer.
investigation depth was fixed to 10 m, equal to the average
depth of the ground improvement. The third day of field Conclusions
survey also used the land streamer and performed one 2-D
production line at a “treated” soil site (i.e. improved From the velocity (Vs) difference map for “treated” (line 3)
ground) (line 3), which ran parallel to the previous and “untreated” (line 1) sites, it seems that MASW surveys
production “untreated” line. Both lines were apart by about are sensitive enough to accurately characterize the relative
30 meters. Dates, survey line number and expected results improvement in subsurface stiffness achieved by the CPR
are summarized in Table 1. Grouting technique. This means that combining field tests
and geophysical surveys will make possible to estimate the
Table 1. Field Survey Summary. settlement reduction factor in the field. This means the
Field Day Survey Line Results efficiency of ground improvement will be directly
correlated to the increase in shear wave velocity.
DAY 1 N/A 1-D Vs test profiles
(Dec. 1, 2015) For this purpose, it is strongly recommended to determine
DAY 2 Line 1 (untreated) 2-D Vs cross section the correlation between MASW results (in both 1-D and 2-
(Dec. 3, 2015) 1-D Vs profile D) and results from conventional in situ test (e.g. SPT,
DAY 3 Line 3 (treated) 2-D Vs cross section CPT, pressuremeter, etc.). The evaluation will have to be
(Dec. 5, 2015) 1-D Vs profile made in layering information (e.g., thicknesses of
embankment and treated zone, and depth to the underlying
stiffer layer, etc.) as well as in stiffness information. This
Results will provide, at least, a first-degree approximation for the
calibration of elastic measurement of shear-wave velocity
Although there were heavy machineries operating nearby (Vs) with respect to the conventional geotechnical
the survey sites during the seismic surveys, an overall parameters used for settlement calculation.
“good to excellent” data quality was obtained from a
combination of using a 10 kg sledge hammer and a big It is also strongly recommended to run MASW surveys at
polymer strike plate, that generated strong surface waves, the same location before and after the ground improvement
especially at low frequencies (e.g., ≤ 10 Hz) that are to produce a velocity difference map similar to the one
essential for moderately deep analysis (e.g., ≥ 5 m). presented in this report. This will provide another evidence
However, shear wave velocity at depths greater than about of effectiveness in using the MASW technique to evaluate
10 m appear to be unstable, possibly due to the relatively the subsurface stiffness variation caused by the CPR
short receiver array length and also due to the extremely Grouting.
low subsurface velocities.
© 2017 SEG Page 5230

Concerning the site conditions, with soft soil deposits, Acknowledgments

embankments and stiff deep strata, the customization of
processing software may be necessary in the future. For The authors would like to thank Engegraut Ltda for
example, a pre-defined layering information can help supporting the research and giving permission to publish
improve the overall accuracy of the inversion results in 1-D this material.
and 2-D analyses.
Figure 3. Velocity (Vs) map at Production Line 1 – untreated soil site.
Figure 4. Velocity (Vs) map at Production Line 3 – improved (grouted) soil site.
Figure 5. Velocity (Vs) difference map obtained by subtracting Line 1 from Line 3. A depth offset was applied to compensate difference in
surface level between the two sites.
© 2017 SEG Page 5231

EDITED REFERENCES
REFERENCES
Almeida, M. S. S., 1998, Site characterization of a lacustrine very soft Rio de Janeiro organic Clay:
Proceedings of the Geotechnical Site Characterization, ISC, 961–966.
Au, S. K. A., K. Soga, M. R. Jafari, M. D. Bolton, and K. Komiya, 2003, Factors affecting long-term
efficiency of compensation grouting in clays: Journal of Geotechnical and Geoenvironmental
Engineering, 129, 254–262, http://doi.org/10.1061/(ASCE)1090-0241(2003)129:3(254).
Baroni, M., 2010, Site investigation in very compressible soft clay deposits in Barra da Tijuca: Master’s
thesis, COPPE/UFRJ.
Briaud, J. L., 1992, The pressuremeter: Taylor & Francis/Balkem.
Cirone, A., 2016, Geotechnical Modelling of CPR Grouting: Master’s thesis, Politecnico di Milano.
Park, C. B., R. D. Miller, and J. Xia, 1999, Multichannel analysis of surface waves: Geophysics, 64, 800–
808, https://doi.org/10.1190/1.1444590.
Park, C. B., and R. D. Miller, 2004, MASW to map shear-wave velocity of soil: Kansas Geological
Survey Open File Report 30.
Riccio, M., M. Baroni, and M. Almeida, 2013, Ground improvement in soft soils in Rio de Janeiro: the
case of the Athletes’ Park: Proceedings of the Institution of Civil Engineers—Civil Engineering,
166, 36–43, https://doi.org/10.1680/cien.13.00008.
Sandroni, S. S., 2006, Obtaining good settlement estimations with very soft soils: the case of Barra da
Tijuca: 13th COBRAMSEG.
Sandroni, S. S., and N. C. Consoli, 2010, About the Brazilian practice with two difficult soils: Extremely
soft and expansible soils: Proceedings of the 14th Brazilian Geotechnical Congress.
Sheriff, R. E., 2002, Encyclopedic dictionary of applied geophysics: (4th ed.): SEG Geophysical
Reference Series No. 13, 429.
Sheriff, R. E., and L. P. Geldart, 1982, Exploration seismology, data-processing and interpretation:
Cambridge University Press.
Yu, H.-S., 2000, Cavity expansion methods in geomechanics: Kluwer Academic Publishers.
© 2017 SEG Page 5232

Seafloor Deformation Monitoring: Past, Present and Future
Paul Hatchell1*, Robin de Vries2, Vicki Gee3, Heather Cousson1, Jorge Lopez1, Shaun Dunn4, Nick Street4,
Adrian Parsons4, Jami Cheramie5 and Eric Fischer5, (1Shell International Exploration and Production Inc.,
2
Shell Global Solutions International B.V., 3Shell Exploration and Production Company, 4Sonardyne
International Ltd., 5Oceaneering Survey Services)
Summary Bourne et al 2006; Sasagawa and Zumberge, 2013;

Dunn et al 2016;)
Over the past decade or so several disparate sensor  Methods that measure relative horizontal displacements
technologies have been gainfully employed to detect the using acoustic ranging between stations (Bourne et al,
very subtle, but measurable, changes at the seafloor. These 2006; Dunn et al, 2016)
technologies include measurement of horizontal and  Absolute horizontal positioning using acoustic ranging
vertical deformations. and GPS (Spiess 1985; Sato et al, 2013; Gagnon et al,
2007)
The ultimate goal of such deformation monitoring systems
is to provide continuous and near real time surveillance of Examples of applying a variety of these methods in
seafloor changes using equipment which is highly sensitive, deepwater fields are discussed together with some
low cost, long life, and most importantly that delivers pros/cons of the different techniques. Following this
significant business value by helping to reduce reservoir summary, we discuss some promising ideas for improving
uncertainty and to identify potential geo-hazards as part of this technology in the future.
a proactive reservoir management philosophy.
Methods using pressure
This paper reviews the past, present and future of seafloor
deformation monitoring starting with early proving trials of Pressure is the most widely used technique for measuring
horizontal and vertical deformation measurement, through seafloor subsidence in deep-water environments.
deployment of equipment at producing reservoirs and Manufacturers of precision pressure sensors quote an
finishing with ongoing research activities targeting the absolute accuracy of 0.01% and a resolution of 0.0001%
amalgamation of seafloor sensors, AUVs and surface which translates into a 10 cm absolute accuracy and 1 mm
drones to provide an integrated real-time deformation resolution at 1000m water depths. In order to take
monitoring system suitable for even the slowest of advantage of the high resolution of these sensors, it is
subsiding reservoirs. important to remove tidal and atmospheric effects from the
measurements and this is most easily accomplished by
Introduction looking at the relative difference of pressures measured at
different locations so that common tidal and atmospheric
Deformation data are useful to constrain reservoir and pressure signals subtract away.
geomechanical models and for understanding issues related
to the stability of wells and seafloor infrastructure. Eiken et al (2000; 2008) developed a technique to measure
the relative vertical height differences of fixed seafloor
In onshore environments, several methods are proven for monuments by moving high precision pressure sensors
detecting surface deformations that use techniques such as between them. By placing reference monuments at
optical leveling, GPS, and InSar (van der Kooij, 1997; locations inside and outside of the expected deformation
Bourne et al, 2006, Allan et al, 2017). These onshore zone they accurately map small changes in subsidence.
methods are accurate to better than cm precision and may Their estimated measurement error is typically a few mm
be closely correlated with dynamic reservoir processes and rivals the precision of methods used onshore.
(Didraga and Lopez, 2016). These techniques are not
applicable offshore however, and new technologies are This method was applied to the Ormen Lange field located
required. offshore Norway where 100+ concrete monuments were
placed around the field as shown in Figure 1 and surveys
Current methods for measuring seafloor deformation in were acquired in 2012 and 2014. The vertical movement
deepwater fields are sorted into three different categories measured between these surveys is displayed in this figure,
based on the type of measurement used: and the maximum subsidence in this 2-year interval is ~ 5.0
 Methods that measure relative vertical displacement cm.
using seafloor water pressure. (Eiken et al, 2000, 2008;
© 2017 SEG Page 5233

Seafloor Deformation Monitoring
The advantage of this method is the high precision <1cm similar to, or even larger than, the subsidence signal
that is obtained, but this method is costly because of the (Chiswell and Lukas, 1989; Chadwick et al, 2006;
extensive use of Remote Operated Vehicles (ROV) Sasagawa and Zumberge, 2013).
required to move the pressure sensors between the
monuments. For this reason, such surveys are made Dunn et al (2016) show long term monitoring of continuous
infrequently and therefore discriminating between long- pressure data also at Ormen Lange using large numbers of
term slow movements and short-term sudden movements pressure sensors that were deployed on the seafloor for
such as due to fault slip (Ottemöller et al, 2005) is not more than 5 years. They observe pressure sensor drift rates
possible. of ~2 cm/yr. Such a drift rate makes it difficult to detect
slowly changing deformations like that in Figure 1, but
works well in fields with larger subsidence rates.
PMT instruments were deployed in a Gulf of Mexico

(GoM) field in water depths ranging from 2500 – 3000 m.
Figure 2 shows the relative pressure differences obtained
from a subset of these sensors located within a few kms of
each other in a region of the field that is thought to subside
very little. The relative pressure difference is determined
by taking the difference between DC-adjusted pressures
and a reference model created by averaging the pressure
data from all the sensors. A 14-day smoothing is applied to
the relative pressure difference to suppress random noise.
This differencing also removes the effects of atmospheric
pressure and ocean current changes.
Long-term stability of the pressure sensors is observed after

the initial start-up drift that may last for ~200 days. The
linear behavior of the pressure differences could be due
either to pressure sensor drift or due to small rates of
subsidence. We observe an upper limit of the drift rate of
these sensors to be 2 cm of water/year.
Figure 1: Subsidence observations in the deepwater Ormen

Lange field using the method of Eiken et al (2000; 2008).
Permanent Seafloor Equipment

Figure 2: Relative pressure differences (displayed in cm of
Bourne et al (2009) describe a method for monitoring water) between several pressure sensors deployed for more
seafloor subsidence using arrays of Pressure Monitoring than 5 years in the GoM.
Transponders (PMTs) permanently installed on the seafloor
that measure pressure every hour for several years. The Due to concerns related to long-term corrosion we placed
main advantages of this approach are that continuous data weight collars on the PMTs to insure against accidental
can be acquired and expensive ROV vessels are not needed separation between the upper measurement system (glass
once the devices are deployed. Data from these devices are sphere) and tripod base as shown in Figure 3. The collars
transmitted back to surface vessels wirelessly using inbuilt were placed after the PMTs had been on the seafloor for
acoustic modems. more than 1 year. Figure 4 shows the relative pressure
The main drawback of employing this method in isolation differences of Figure 2 after de-trending the data in a 200-
is that pressure sensors have long-term drifts that can be day interval after the weight collars were place to remove
© 2017 SEG Page 5234

linear subsidence or drift features. We clearly observe the performance and usefully identifies problematic sensors.
small sudden changes in pressures associated with less than This method can also be applied post-retrieval and we are
10 mm movements, or tilts, that occurred when the weight testing this on sensors that were recovered after a 5.5 year
collars were placed. The resolution for detecting these rapid deep-water deployment to compare against the long-term
movements is estimated to be ~2.0 mm. drift rates observed on the seafloor.
Figure 3: Weight collar being installed on an PMT device.
Figure 5: 120 pressure sensors held at the same

temperature and pressure for months enabling drift
characteristics to be determined and badly behaving sensors
to be detected.
Horizontal Deformation Monitoring
The PMT devices described by Bourne et al (2009) are

used to measure horizontal seafloor deformations. The
devices frequently observe acoustic travel times between
pairs of sensors and the sound velocity of the water. These
Figure 4: Sudden change in pressures that occurred when two observables are combined to create a time series of
weight collars were place on the PMT instruments. distance variation between every PMT pair.
Correcting Pressure Measurements: Geodetic network adjustment is performed to translate time
series of distance into time series of station (x,y) positions.
Sasagawa and Zumberge (2013) describe a “Self- Furthermore, this step allows for the rejection of
Calibrating Pressure Recorder” or SCPR that measures the measurement errors, i.e. measurements that do not ‘fit’ the
long-term drift of pressure sensors by incorporating a geometric constraints can be removed.
precision Dead Weight Tester (DWT) within the hardware
deployed on the seafloor. Periodic calibration of the The errors associated with this method depend on the
pressure reading against the DWT allows the sensor drift spatio-temporal variability of the sound velocity,
correction to be estimated and removed from the data. instrument stability, and the spacing of the PMT pairs. To
Accuracy of this system is expected to reduce the errors to remove gross errors before network adjustment,
less than 1.0 cm/year. Shell deployed several of these multivariate statistical testing is performed on travel time
instruments in two GoM fields and had difficulties with reciprocals (i.e. from station A-B and B-A), combined with
long-term reliability of these complex electromechanical the sound velocity data, which behave under a multivariate
systems distribution. To detect and correct instrumental drifts, the
individual time series are tested against a sound velocity
An alternative method for estimating pressure drift model based on the entire population of sensors, assuming
correction is to carry out long-term measurements in the the network is stable on average (de Vries, 2015).
laboratory prior to deployment of the instruments on the
seafloor (Figure 5). Provided that the drifts are largely Such deployment and method have been applied at Ormen
linear and predictable, this method provides a means to Lange from 2011 - 2016 (Dunn et al, 2016). Results are
correct them. Prior to a recent deployment of PMT-style displayed in Figure 6, which shows modeled (right) versus
devices we placed pressure sensors at temperatures and measured (left) changes in distance for Ormen Lange.
pressures expected in the field and measured the sensor Changes of distance of 5 PPM per year (i.e. 5mm change
drift over a period of a few months. This also on a 1km distance) can be detected with strong
comprehensively pre-screens the sensors for quality and
© 2017 SEG Page 5235

significance. More recently, ranging networks have been accuracy by using the PMT devices as a long-baseline
deployed at two Gulf of Mexico fields. (LBL) acoustic array during bathymetric surveys.
Historically, data from seabed PMTs were extracted using

acoustic modems deployed from manned surface vessels
with a commensurately high HSE exposure and significant
cost implications. It is now standard practice to send
unmanned drones, such as the Liquid Robotics
WaveGlider, to collect the data and transport it via satellite
communications directly to shore.
During future data harvesting missions these drones will

Figure 6: Measured (left) vs. modeled (right) strain rates at also be used to perform GPS-Acoustic measurements
Ormen Lange. Each line represents a ranging arc between (Spiess, 1985; Sato et al, 2013; Gagnon et al, 2007) of the
two PMT’s. Green denotes no horizontal displacement, depth (z) and horizontal (x-y) displacement of PMTs in
blue shows extension and red shows contraction. both relative and absolution position terms providing
another means to recalibrate the array towards <1cm
Future Trends: precision.
Several ideas are in development that will improve our Summary

sensitivity and lower the cost of seafloor monitoring. Laser
ranging using Autonomous Underwater Vehicles (AUV) Much has been learned about seafloor deformation
can accurately measure the seafloor bathymetry while also monitoring technology from trials and deployments in
estimating PMT depth using pressure sensors and the laser- Norway (Ormen Lange) and the Gulf of Mexico.
determined offset distance from the AUV. Based on current Campaign-style pressure surveys provide reliable
subsea laser technology, sub-centimeter ranging resolution subsidence estimates, but cannot capture nonlinear and
is expected. Surveys of the type carried out by Eiken et al discontinuous deformation regimes.
(2000; 2008) could be accomplished using the AUV to
move a precision pressure sensor between monuments or This capability gap is filled by the deployment of
simply seafloor features of interest to measure relative continuous pressure and acoustic monitoring networks,
depths between these features accurately. Repeating this which are able to detect rapid changes in depth as precisely
pressure sensor recalibration survey periodically will as 2mm. Instrumental drift (up to 2cm/yr) poses a problem
permit more accurate subsidence monitoring. Figure 7 for precise deformation rates (especially for subtly
shows an example of the detail that is captured via laser subsiding fields) and has sparked the development of
scanning of a PMT on the seafloor. various remedies: Self-Calibrating Pressure Recorders and
lab-calibration of large batches of sensors to select
predictably-behaving instruments. Together with improved
instrument designs, this has brightened the outlook for
performance of continuous monitoring networks.
On the near future’s horizon are various novel technologies

that have the potential to reduce the cost of campaign-style
subsidence surveys and re-calibrate the sensors on
continuous monitoring networks. Here, we mention AUV’s
that can perform laser ranging and pressure surveys, as well
as precise GPS-Acoustic positioning of seafloor nodes by a
WaveGlider. Improved precision will make these methods
appropriate for many more fields across the world.
Figure 7: Laser scan of a 3-m high PMT device compared Acknowledgements

to a drawing of the instrument (inset) The Authors would like to acknowledge Shell E&P,
Chevron USA, BP E&P, and the Ormen Lange asset
The use of an AUV will significantly shorten the duration partners (A/S Norske Shell, Petoro AS, Statoil Petroleum
and costs of this survey when compared to using an ROV, AS, DONG E&P AS, and ExxonMobil) for permission to
and has the added benefit of improved AUV positional publish this paper.
© 2017 SEG Page 5236

EDITED REFERENCES
REFERENCES
Allan, M. E., P. B. Leezenberg, and R.F. Hanssen, 2017, InSAR — Pro-active remote sensing for
reservoir management and monitoring environmental safety: 79th Annual International
Conference and Exhibition, EAGE, Extended Abstracts, Th PRM 01,
http://doi.org/10.3997/2214-4609.201700027.
Bourne, S. J., Hatchell, P. J., Partridge S. C., Leaf, C. J., Klemm, H., Kampshoff, S., and Cook, A. M.,
2009, An autonomous seafloor system for monitoring reservoir deformation: 71th Annual
International Conference and Exhibition, EAGE, Extended Abstracts, Z034,
http://doi.org/10.3997/2214-4609.201400567.
Bourne, S. J., Maron, K., Oates, S. J., Mueller, G., 2006, Monitoring reservoir deformation on land —
Evidence for fault reactivation from microseismic, InSAR, and GPS data: 68th Annual
International Conference and Exhibition, EAGE, Extended Abstracts, E026,
http://doi.org/10.3997/2214-4609.201402272.
Chadwick, W. W., S. L. Nooner, M. A. Zumberge, R. W. Embley, and C. G. Fox (2006), Vertical
deformation monitoring at Axial Seamount since its 1998 eruption using deep-sea pressure
sensors, Journal of Volcanology and Geothermal Research, 150, 313–327,
http://doi.org/10.1016/j.jvolgeores.2005.07.006.
Chiswell, S. M., and R. Lukas (1989), The low-frequency drift of Paroscientific pressure transducers,
Journal of Atmospheric and Oceanic Technology, 6, 389 – 395, http://doi.org/10.1175/1520-
0426(1989)006<0389:TLFDOP>2.0.CO;2.
de Vries, R. (2015), Impact of instrumental drift on the estimation of deformation parameters from an
acoustic ranging network on the seabed above the Ormen Lange gas field: M.sc. Thesis.
Didraga, C., and Lopez, J., 2016, Correlating frequent InSAR deformations with reservoir pressure for
areal conformance in thermal EOR at Peace River: 78th Annual International Conference and
Exhibition, EAGE, Extended Abstracts, Tu LHR2 07, http://doi.org/10.3997/2214-
4609.201600578.
Dunn, S, P. Hatchell, A. van den Beukel, R. de Vries and T. Frafjord, 2016, A long-term seafloor
deformation monitoring campaign at Ormen Lange gas field: First Break, 34, 55–64.
Eiken, O., T. Stenvold, M. Zumberge, H. Alnes, and G. Sasagawa, 2008, Gravimetric monitoring of gas
production from the Troll field, Geophysics 73, no. 6, WA149–WA154,
https://doi.org/10.1190/1.2978166.
Eiken, O., M. Zumberge, and G. Sasagawa, 2000, Gravity monitoring of offshore gas reservoirs 70th
https://doi.org/10.1190/1.1816087.
Gagnon, K. L, and C. D. Chadwell, 2007, Relocation of a seafloor transponder — Sustaining the GPS-
Acoustic technique: Earth Planets and Space, 59, 327–336.
Ottemöller, L., H. H. Nielsen, K. Atakan, J. Braunmiller, and J. Havskov, 2005, The 7 May 2001 induced
seismic event in the Ekofisk oil field, North Sea, Journal of Geophysical Research, 110, B10301,
http://doi.org/10.1029/2004JB003374.
Sasagawa, G., and M. A. Zumberge, 2013, A self-calibrating pressure recorder for detecting seafloor
height change: IEEE Journal of Oceanic Engineering, 38, 447–454,
http://doi.org/10.1109/JOE.2012.2233312.
© 2017 SEG Page 5237

Sato, M., M. Fujita, Y. Matsumoto, H. Saito, T. Ishikawa, and T. Asakura, 2013, Improvement of
GPS/acoustic seafloor positioning precision through controlling the ship’s track line: Journal of
Geodesy, 87, 825–842, http://doi.org/10.1007/s00190-013-0649-9.
Spiess, F. N., 1985, Suboceanic geodetic measurements: IEEE Transactions on Geoscience and Remote
Sensing, GE-23, 502–510, http://doi.org/10.1109/TGRS.1985.289441.

van der Kooij, M., 1997, Land subsidence measurements at the Belridge oil fields from ERS InSAR Data:
ESA, Atlantis Scientific, Inc..
© 2017 SEG Page 5238

Marine Electromagnetics – Present and Future
Leonard J. Srnka* and Steven Constable, Institute of Geophysics and Planetary Physics, Scripps Institution of
Oceanography, University of California San Diego, La Jolla, CA 92039-0225
Summary
Since industry began funding marine electromagnetics
(EM) in the 1980s, tremendous progress has been made in
applying these methods not only to hydrocarbon
exploration but also to a wide range of applications. After
more than two decades of excitement and investment, many
successes and some failures, marine EM use has now
greatly diminished due to a combination of factors,
including resolution and depth limitations, imaging
uncertainties, perceived high costs, difficulties in business
integration, and the recent down cycle in hydrocarbon
markets. Fortunately, acquisition costs are declining due to Figure 1: Schematic of marine MT (natural-source) and
the arrival of continuously towed systems, subsurface CSEM (controlled source) EM methods. The multi-
depths of investigation are increasing, and cooperative component electric and magnetic seafloor receivers are
seismic-EM inversions are beginning to produce better essentially the same for both source types.
estimates of lithologies and fluids, including for marine
hydrates. Enabled by the technical progress made in deep background noises and the direct source airwave arrival in
water for hydrocarbon use, offshore EM is expanding to CSEM. That need has been met in many instances, fulfilled
other applications. Virtually any EM onshore application is primarily by new and established geophysical contactors
now a candidate for the offshore. These include permafrost, who leveraged research from a few key universities,
geothermal, minerals, contaminant mapping, and including UCSD in the USA and Southampton/NOC in the
groundwater investigations. Offshore EM methods are now UK, as well as providing advancements of their own. These
firmly installed in the geophysical toolkit, and their uses contractors collectively saw the potential, and then took on
will grow. considerable technical and business risks.
Introduction
Industry embarked on a journey decades ago to remotely

measure subsurface electrical properties offshore. Figure 1
depicts the basic acquisition methods for marine EM,
which uses both natural and controlled sources.
Magnetotellurics (MT) for industry began in 1996, testing
the mapping of Gulf of Mexico salt structures at the Gemini
prospect (Hoversten at al, 2000). Other large-scale
geological applications (Figure 2) include exploring the
structure and state of mid-ocean ridges, and studying
ocean-continent subduction zones (Naif et al., 2015). But
the primary oil and gas business driver was to determine
rock and fluid properties better than could be done with
reflection seismology alone. Marine controlled source
electromagnetics (CSEM) was developed to address that Figure 2: Three established uses of marine EM: exploring
need. CSEM arrived in 2001 offshore Angola, focused on the structure and thermal state of mid-ocean ridges (left),
detection of reservoired hydrocarbons (Eidesmo et al, 2002; studying ocean-continent subduction zones (top), and
Srnka et al., 2006; Constable and Srnka, 2007). It was well exploring for hydrocarbons (bottom).
known previously from first principles that CSEM would
only determine subsurface resistivity, and not be a new In the case of CSEM, which is the most appropriate tool for
direct detection tool. fluid estimation since it is primarily sensitive to the vertical
component of subsurface resistivity, pre-drill prediction
There is no maximum depth limitation to these methods, accuracy of hydrocarbon presence in siliciclastic settings
but shallow water (<50 meters) can be challenging due to rivals that of seismic DHIs, within the CSEM depth range
of applicability of roughly 2500 meters below the sea floor,
© 2017 SEG Page 5239

Marine EM - Present and Future
depending on the average background vertical resistivity and weaknesses of the technology are better understood and
(Constable and Srnka 2007). Effective depth of its value proposition becomes clearer.
investigation and resolution both improve as the average

resistivity of the section increases, but concomitantly the Business Paradigms and Challenges
likely resistivity contrast between hydrocarbon-bearing
units and other strata decreases, thus decreasing CSEM CSEM users faced a range of business challenges that seem
sensitivity to resistive targets since to first order the to have been only partly overcome. These challenges were
maximum response is proportional to the net thickness- not unexpected for a major new technology appearing in a
resistivity contrast product of the target. very large, mature, and fairly conservative industry. This is
especially true since CSEM has too often been viewed as a
Some false positives for hydrocarbon detection have been competitor for well-established seismic reflection
encountered, but less than a dozen have been reported out techniques, rather than as a complementary method. As
of the many hundreds of surveys used in pre-drill many authors have said in presentations and publications,
predictions that were correct, either wet or dry holes. integration of CSEM with seismic and other geoscience
Perhaps most important are the negative (i.e dry hole) data is essential for effective use of this relatively new
results: there have been virtually no false negatives of technology. But few geophysical data interpretation
material size reported (Hesthammer et al. 2012, 2015; see packages integrate seismic reflection and CSEM data in an
Figure 3). An apparent false negative in an early CSEM organic way, although progress has been made recently.
application over a major West Africa discovery has been
shown to be due to faulty survey design and inadequate Obviously, there are significant business motivations to
data analysis. This strong CSEM technical ability for honor drill well commitments in licenses, in spite of
vetting probable dry hole locations seems to have been negative CSEM responses, rather than paying a back-out
under-utilized by industry. fee. Not the least of these is gaining geological knowledge
of the basin by drilling. Another rationale for down-
weighting negative CSEM responses is mistrust of the
results, perhaps due to the low spatial resolution or
uncertainties in determining resistivity values, rather than
just the newness of the technology. But because of the
much higher sensitivity to a range of hydrocarbon
saturations than in the case of reflection seismic (e.g.
Archie’s law), especially where low gas saturations can
give false seismic DHIs or uncertain fluid edges, such
mistrust deserves more thought. Perhaps a re-think of
exploration business models is also appropriate to better
value well avoidance, especially in times of low
Figure 3: CSEM prediction success rate for wells in the hydrocarbon prices where controlling exploration costs is
Barents and North Sea basins. All of the “mostly dry” wells paramount.
had weak or no CSEM responses (NAR=normalized
amplitude response above the so-called “background”, a A global market factor that may be playing a role in CSEM
parameter favored by some CSEM contractors; an NAR of use, as well as for other geophysical exploration tools, is
0.15 was assumed to be the minimum reliable signal). the changing concept of “peak oil”. Originally defined as
the maximum supply capacity, it has come to mean
Business adoption for routine exploration operations has maximum demand. In this second interpretation, priorities
been uneven since CSEM commercialization in 2001, with of both commercial and national oil companies may shift to
a group of early adopters such as ExxonMobil, Shell, and a more measured program of discovery and development,
Chevron investing heavily and then scaling back. Some valuing long-term technical strengths, local resources and
independents and national oil companies picked up the security of supply over near-term market economics. For
technology later, and some have continued its use at example, there is a large and growing EM methods effort in
various levels of application. CSEM research and China, including marine EM research, where developing
development and application expenditures have followed and sustaining a national capability appear to be valued.
the classic “S curve” characteristic of emerging
technologies: rapid growth to a peak after early successes, a Recent Technical Advances
long decline after prediction failures, a bottoming-out, and
then slow recovery to a sustainable level as the strengths Continuously towed CSEM systems have appeared that
promise to substantially reduce cost and to expand the
© 2017 SEG Page 5240

applications envelope. EM streamers containing single- Opportunities for New Kinds of Exploration
component (inline E-field) sensors can now be towed
separately or together with 2D seismic streamers from As mainstream oil and gas exploration applications have
dedicated vessels at 4 to 6 knots, rather than at the 1 to 2 waned, other types of offshore resources are coming into
play as viable marine EM targets. These include: marine
hydrates for possible gas supplies, for example in the Sea
of Japan (Figure 6) and other areas of the Far East; for
marine geotechnical use such as drilling and infrastructure
hazard assessment, and also for studies of ancient and
current climate change (Figure 7).
Figure 4: Simultaneous EM and seismic streamer

acquisition (image courtesy of PGS)
knots characteristic of deep-towed CSEM (Figure 4). It is

reported that cross-talk between the coterminous seismic Figure 6: Seismic multibeam (top left) and CSEM
and EM data is minimal in this system. Much larger electric resistivity (bottom left), Sea of Japan. Image courtesy of
source dipole moments are becoming available, giving the Japan Research Consortium for Methane Hydrate
considerably greater geological depth of investigation. In Resources.
addition, three component EM-only receiver systems now
exist that can be towed at any depth using small vessels of
opportunity, such as the Scripps Vulcan system (Figure 5),
adding cross-line and vertical E-field data that are valuable
in edge detection and in data inversions especially when
electrical anisotropy is present.
Figure 5. UCSD Scripps Vulcan II marine EM system.

Streamer lengths up to 1 km have been achieved.
In addition to advances in acquisition, much progress has Figure 7: Inverted 2D resistivity from CSEM (top), and
been made in EM data interpretation and imaging.
seismic P-wave velocity (bottom) at Hydrate Ridge,
Although true joint nonlinear inversion of EM and seismic Oregon (Weitemeyer et al, 2011). Not all low velocity
data has not been successfully accomplished, due to the zones correspond to hydrates. Free gas shows as highly
very different scale lengths of the two types of data, the resistive areas, indicating high gas saturation.
much larger non-uniqueness of EM imaging, and the large
amount of required computation, seismically guided EM Emerging marine EM exploration applications also include
inversion is now practical. The best results can be achieved permafrost (Sherman et al., 2016), geothermal, and
when there is well control and calibrated rock physics minerals work using both conventional and newly
models for the area (Alvarez et al., 2017).
developed low-power EM seafloor sources. Environmental
applications for mapping contaminants and for fresh
ground water supplies are now viable offshore, and
research surveys are progressing. Whether CSEM in any of
© 2017 SEG Page 5241

its modes is a useful method for time-lapse (4D) studies,

either for hydrocarbon depletion work or for monitoring
gas storage (hydrocarbons or sequestered CO2) continues

to be a subject for research (Orange et al, 2009). In all of
these cases, modern computing is a major enabler, without
which today’s applications would be largely impossible.
Conclusions
Although the origin of marine electromagnetic geophysical
methods can be traced to DC surveys offshore Cornwall in
the early 1920s, the marine methods we know today in
industry are newcomers compared to seismic, gravity, and
magnetic methods. A great deal of progress has been made
in less than three decades, especially considering the
relatively small investment in EM compared to seismic.
Despite the current downturn, marine EM will move ahead
technically and commercially, and will continue to offer
viable geophysical options for geoscience applications.
Acknowledgements
Many thanks to our fellow members of the global marine
EM community for their discussions, suggestions, and
contribution for this talk, including the members of the
UCSD Scripps Seafloor Electromagnetics Consortium,
especially RSI, PGS, EMGS, and Statoil.
© 2017 SEG Page 5242

EDITED REFERENCES
REFERENCES
Alvarez, P., A. Alvarez, L. MacGregor, F. Bolivar, R. Keirstead, and T. Martin, 2017, Reservoir
properties prediction integrating controlled-source electromagnetic, prestack seismic, and well-
log data using a rock-physics framework: Case study in the Hoop Area, Barents Sea, Norway:
Interpretation, 5, SE43–SE60, http://doi.org/10.1190/INT-2016-0097.1.
Constable, S. and L. J. Srnka, 2007, An introduction to marine controlled-source electromagnetic methods
for hydrocarbon exploration: Geophysics, 72, No. 2, WA3–WA12,
http://doi.org/10.1190/1.2432483.
Eidesmo, T., S. Ellingsrud, L. M. MacGregor, S. Constable, M. C. Sinha, S. E. Johansen, F. N. Kong, and
H. Westerdahl, 2002, Sea Bed Logging (SBL), a new method for remote and direct identification
of hydrocarbon filled layers in deepwater areas: First Break, 20, 144–152,
http://doi.org/10.1046/j.1365-2397.2002.00264.x.
Hesthammer, J., A. Stefatos, and S. Sperrevik, 2012, CSEM efficiency — Evaluation of recent drilling
results: First Break, 30, 47–55, http://doi.org/10.3997/1365-2397.2012007.
Hesthammer, J., 2015, Private communication.
Hoversten, G. M., S. C. Constable, and H. F. Morrison, 2000, Marine magnetotellurics for base-of-salt
mapping: Gulf of Mexico field test at the Gemini structure: Geophysics, 65, 1476–1488,
http://doi.org/10.1190/1.1444836.
Naif, S., K. Key, S. Constable, and R. L. Evans, 2015, Water-rich bending faults at the Middle America
Trench: Geochemistry, Geophysics, Geosystems, 16, 2582–2597,
http://doi.org/10.1002/2015GC005927.
Orange, A., K. Key, and S. Constable, 2009, The feasibility of reservoir monitoring using time-lapse
marine CSEM: Geophysics, 74, No. 2, F21–F29, http://doi.org/10.1190/1.3059600.
Sherman, D., P. Kannberg, S. Constable, 2017, Surface towed electromagnetic system for mapping of
subsea Arctic permafrost, Earth and Planetary Science Letters, 460, 97–104,
http://doi.org/10.1016/j.epsl.2016.12.002.
Srnka, L. J., J. J. Carazzone, M. S. Ephron, and E. A. Eriksen, 2006, Remote reservoir resistivity
mapping: The Leading Edge, 25, 972–975, http://doi.org/10.1190/1.2335169.
Weitemeyer, K. A., S. Constable, and A. M. Tréhu, 2011, A marine electromagnetic survey to detect gas
hydrate at Hydrate Ridge, Oregon: Geophysical Journal International, 187, 45–62,
http://doi.org/10.1111/j.1365-246X.2011.05105.
© 2017 SEG Page 5243

Towards full-bandwidth inversion in complex geologies
Uwe Albertin*, Chevron Energy Technology Company
Summary techniques include but are not limited to: 1. Traveltime FWI,
Over the last decade, the understanding of waveform where conventional FWI residual is replaced by a time shift
inversion technology and its application to seismic data on a obtained from correlation or data-warping analysis, and the
large scale has progressed significantly. Here I discuss the shift is related to the derivative or phase and envelope of the
concepts behind some of the techniques being developed in residual (Luo and Schuster, 1991, Ma and Hale, 2013, Kun
the general area of wavefield-based inversion, and present a et al, 2015), 2. Envelope FWI, which separates the data into
number of examples that illustrate the directions this instantaneous phase and envelope, and uses the envelope in
technology is taking towards becoming a full bandwidth the residual (Wu et al, 2014), 3. Spectral balancing and
multi-parameter seismic inversion in complex areas. deconvolution methods to extend the low frequency end of
the FWI residual (Lazaratos et al, 2011, Fei et al, 2012), 4.
Classic acoustic full waveform inversion in shallow Weighted match filter design, where a filter is designed to
complex geologies and acoustic impedance inversion match modeled data with acquired data, after which
Numerous examples of successful application of standard components of the filter are weighted so the data match is no
acoustic full waveform inversion (FWI) for shallow complex longer perfect, but not quite cycle skipped (Warner and
geologies have been presented over the past decade (Sirgue Guasch, 2016, D. Sun et al, 2016), 5. Optimal transport, in
et al, 2010). Standard acoustic FWI tries to minimize the which a weighting function is constructed through inverse
objective function Laplace-type filters that represents the minimum work
𝐽 = (1/2)|ℱ(𝑚) − 𝑑|2 required to change the modeled data into the acquired data,
where 𝑚 is the model, 𝑑 is full-bandwidth data, and ℱ is the and the weighting function is used as the residual (L.
forward modeling operator. For shallow velocity recovery in Metevier et al, 2016). One of the primary advantages of all
media with a positive velocity gradient, the diving wave of these techniques is that most of the standard FWI
component of the data is quite effective. In this regime a functionality, and in particular the gradient computation, is
point response in the gradient forms a banana shape, leading largely unchanged in implementation of the methods; only
to long wavelength updates to the maximum diving wave the residual is changed and the objective function is
depth. In production practice, actual fitting of the data is modified. Hence computational performance of the
typically not done, but some amplitude mitigation is used, algorithm is largely unaffected when using these techniques.
such as the use of phase in frequency domain inversion. The techniques that appear to be most successful in
Pushing the frequency content of the inversion higher production settings are methods based on correlation and
and moving deeper in the section, standard FWI relies more warping, since there is quite a bit of experience in service
on reflections in the data, and if there is sufficient bandwidth organizations in correlation-based picking and gather
separation between the background model and the model warping stemming from development of standard ray-based
updates being inverted for, the inversion resembles least- tomographic methods and stack optimization methods.
squares migration or impedance inversion (Routh et al,
2017). The point response behavior of the inversion gradient The tomographic operator and tomographic extensions
in this regime is the same as the kinematics of a classic smile Since its early development, it has been known that the FWI
impulse response in seismic migration. In this regime it is gradient from reflection data contains both long and short
beneficial to incorporate density, even if the density-velocity wavelength components (Mora, 1989); the short wavelength
ratio is treated heuristically. An example of a higher- component comes from the crosscorrelation of primary
frequency acoustic FWI is shown in Figure 1. events in the gradient imaging condition, similar to a
migration. The long wavelength tomographic component, or
Long-wavelength extensions via data conditioning in the so-called rabbit-ear component, comes from the correlation
data domain of backscattered energy with primary energy in the gradient
There are two primary issues with conventional FWI that imaging condition. Unfortunately, in a standard waveform
cause issues for model updating. The first is cycle skipping, inversion, backscattered energy appears only after relatively
in which the modeled data and the background data are out higher frequency reflection events are present in the model,
of phase by more than half a cycle, while the second is the positions of which changes as the long wavelength
insufficient bandwidth to reconstruct a long-wavelength update is made to the model. This dilemma is the
model change at the maximum opening angle available at the generalization of the chicken-and-egg problem in standard
anomaly. Mitigation of these issues has led to a tremendous imaging, where an optimal stacked image cannot be obtained
number of extensions to FWI, many of which try to extend without a good background model, and tomography cannot
performance in long wavelength recovery by lowering the be done without a good image of gathers. In general it is
effective frequency content of the data residual. These quite difficult for standard FWI to shift the position of
© 2017 SEG Page 5244

Towards full-bandwidth inversion
reflectors once they are inserted into the model; attempts to angle. Despite difficulties associated with phase behavior of
do so tend to require a very large number of small model the standard DSO gradient (Fei and Williamson, 2011), good
changes. In addition, the tomographic component of results can be obtained on real data as seen in Figure 2. The
standard FWI tends to be weak. Methods that try to method also benefits from the use of augmented techniques
accomplish long wavelength updating using the standard such as data warping, for which no nonlinear objective
correlation imaging condition in the gradient usually employ function exists (Shen and Symes, 2015).
some method of separation of the tomographic component The tomographic operator arises naturally in FWI in
from the high-frequency-reflector component, and some a number of ways. If one assumes a scale separation between
have been successful (Tang et al 2013). It can be beneficial longer wavelength background models and shorter
for such schemes to use more parameters in the model; in wavelength model perturbations associated with reflectors
particular the separation is more effective using parameters so that 𝑚 = 𝑚0 + 𝛿𝑚, and one examines the linearized
such as impedance or density for the high-frequency- FWI, or least-squares RTM objective function
reflector component (Brossier et al 2013), and may benefit 1
𝐽 = ( ) |𝐹𝛿𝑚 − 𝛿𝑑|2
if the analysis is done in pseudotime (Plessix et al 2012). 2
Once the long-wavelength component has been updated, then variation of this objective function with respect to the
new reflectors can be generated. background model, assuming 𝛿𝑚 is independent of 𝑚0 ,
gives the gradient with respect to 𝑚0 as 𝐷𝐽 =
In order to gain more control over the tomographic 𝐷𝐹 ∗ [𝛿𝑚, 𝐹𝛿𝑚 − 𝛿𝑑], which corresponds to the
component, a large amount of research over the past fifteen tomographic reflection inversion of Xu et al (2012). In the
years in tomographic wavefield methods has been devoted simplest case, 𝛿𝑚 might be chosen to be 𝐹 ∗ 𝛿𝑑, the migrated
to the explicit computation of the correlation of image itself. Because tomographic reflection inversion uses
backscattered energy with primary energy for some the linearized FWI residual, it is susceptible to cycle
appropriate residual field. Central to almost all schemes is skipping in the same manner as FWI. However, because it
the tomographic operator. Let 𝐷ℱ𝛿𝑚 = 𝐹𝛿𝑚 = 𝛿𝑑 be the uses this same residual, any method that avoids cycle
directional derivative of the forward modeling operator skipping in the data domain can in principle be used in the
producing a band-limited data perturbation 𝛿𝑑, which may tomographic operator; in particular all of the methods 1-5
be identified with Born modeling, so migration becomes listed in the previous section can in principle lead to
𝛿𝑚 = 𝐹 ∗ 𝛿𝑑 The tomographic operator is related to the residuals used in tomographic updates (Sun et al, 2016).
derivative of the Born scattering operator, which can be One of the difficult issues for wavefield tomographic
written 𝐷[𝐷ℱ𝛿𝑚]𝛿𝑚 = 𝐷𝐹[𝛿𝑚, 𝛿𝑚]. Since this operator methods is the separation of illumination effects from
produces data, we can form the inner product of the result of velocity effects in constructing the residuals used in the
this operator with an arbitrary data set 𝑑′ by integrating over backprojection. In complex areas, methods that achieve this
source, receiver, and time coordinates. Then the second separation may have some advantage. Consider, for
order scattering operator is related to the tomographic example, the image-domain linearized FWI objective
operator through the relation function:
< 𝐷𝐹[𝛿𝑚, 𝛿𝑚], 𝑑′ >=< 𝛿𝑚, 𝐷𝐹 ∗ [𝛿𝑚, 𝑑′] > 1
𝐽 = ( ) |𝐹 ∗ (𝐹𝛿𝑚 − 𝛿𝑑)|2
Here 𝐷𝐹 ∗is the derivative of the migration operator with 2
respect to the model, which involves the correlation of which has a residual that corresponds to migration of the
source and receiver wavefields with Born-scattered linearized FWI residual. Neglecting the variation of the
wavefields thereby computing the rabbit-ear. normal operator 𝐹 ∗ 𝐹 with respect to the background model,
Some of the earliest uses of this operator in wavefield this objective function leads to the gradient
methods were in the development of wave-equation 𝐷𝐽 = 𝐷𝐹 ∗ [𝐹 ∗ (𝐹𝛿𝑚 − 𝛿𝑑), −𝛿𝑑]
migration velocity analysis (WEMVA) (Biondi and Sava, which is essentially a slight variation of image-domain FWI
1999, Shen and Symes, 2003), with the objective function advocated by Zhang and Schuster (2013). This method has
1 built into it a mechanism for separating illumination effects
𝐽 = ( ) |𝐴𝐹 ∗ 𝛿𝑑|2 from velocity effects because of the residual 𝐹 ∗ 𝐹𝛿𝑚 −
2
where 𝐹 ∗ 𝛿𝑑 is the migrated image in an extended image 𝐹 ∗ 𝛿𝑑. In the case of subsurface angle gathers, if 𝛿𝑚 here is
domain, typically subsurface offset, and 𝐴 is an annihilation augmented with an enhancing operator so that it corresponds
operator that removes energy from the image that is roughly to a stacked image, then 𝐹 ∗ 𝐹𝛿𝑚 will be a flat angle
consistent with a correct background model. Variation of gather that has nothing but illumination effects in it.
this objective function with respect to the background gives Subtracting this from the migrated image gather
the gradient 𝐷𝐽 = 𝐷𝐹 ∗ [𝐴∗ 𝐴𝐹 ∗ 𝛿𝑑, 𝛿𝑑]. In the case of 𝐹 ∗ 𝛿𝑑 should lead to a residual that is more a measure of
differential semblance in subsurface offset (DSO), the velocity error. An example of this is shown in Figure 3.
operator 𝐴 corresponds to multiplication by offset, which Cycle skipping in tomographic FWI can also be
corresponds to differentiation across angle in subsurface avoided by augmenting the linearized FWI objective
© 2017 SEG Page 5245

function with a term that penalizes energy associated with an Moving to the elastic realm
incorrect velocity in the image Because of the difficulties associated with elastic attribute
1 recovery from conventional imaging in complex areas, there

𝐽 = ( ) |𝐹𝛿𝑚 − 𝛿𝑑|2 + 𝜆2 |𝐴𝛿𝑚|2
2 has recently been increased interest in elastic FWI for the
(Sun and Symes, 2012, Biondi and Almomin, 2012). For the recovery of these parameters. To relate to AVO analyses
case where the image 𝛿𝑚 is a subsurface offset gather, the and mechanical earth models used in well planning,
annihilator eliminates focused energy, so the penalty term orthorhombic and tilted-orthorhombic symmetries appear to
drives the optimization towards increased focusing and strike a good balance between parameters of interest in
flatter gathers in the angle domain. Symes and coworkers subsurface integrity analysis, and complexity of the FWI
have analyzed techniques for optimization of this objective algorithm (Albertin et al 2016, Alkhalifah et al 2016). For
function using the variable projection method, which solves orthorhombic symmetry, the wave propagation system may
for 𝛿𝑚 first using the normal equations (𝐹 ∗ 𝐹 + be written
𝜆2 𝐴∗ 𝐴)𝛿𝑚 = 𝐹 ∗ 𝛿𝑑 and then solves for the update in the
background model by computing the background-model
gradient of the reduced object function. Ideas along these where
lines have been investigated in numerous domains including
subsurface offset, subsurface-time offset, surface offset,
common-shot, and extended source domains, and
improvements in phase behavior over standard WEMVA
DSO have been noted.
More recently there is increased interest in the
tomographic operator as part of the FWI Hessian, which in
principle can be incorporated into a Newton optimization.
To this point I have considered 𝛿𝑚 as being associated with
the high frequency part of the model update. Suppose I relax
that condition and consider 𝛿𝑚 to be a general model
change. Expanding the FWI objective function to second
order gives
1 Here 𝑝 is the wavefield, 𝑓 is the source, 𝐶 are the stiffnesses,
𝐽(𝑚 + 𝛿𝑚) = 𝐽(𝑚)+< 𝐹𝛿𝑚, ℱ(𝑚) − 𝑑 > + ( ) 𝑉 are the particle velocities, and 𝜎 are the stresses. The
2
1 objective function for elastic FWI from hydrophone data is
< 𝐹𝛿𝑚, 𝐹𝛿𝑚 > + ( ) then
2 1
< 𝐷𝐹𝛿𝑚𝛿𝑚, ℱ(𝑚) − 𝑑 > 𝐽 = ( ) |𝑆𝑝 − 𝑑|2
From this one may recognize the Hessian acting on the 2
model change as 𝐹 ∗ 𝐹𝛿𝑚 + 𝐷𝐹 ∗ [𝛿𝑚, ℱ(𝑚) − 𝑑], which where 𝑆 is a projection operator that projects onto the
involves the tomographic operator in a natural way. hydrophone component. Recovery of elastic parameters in
Separation into short and long wavelength parts of the such an inversion relies on careful amplitude matching
update and inversion of the Hessian should lead to improved between modeled and actual data, so the inversion is using
long and short wavelength updating, and it will be interesting the same amplitude vs. offset information that drives
to see how these algorithms perform. standard AVO inversion. Results on synthetic data in 3D
Of the large number of options available for indicate good recovery of shear and compressional velocities
application of the tomographic operator, methods that are below salt of moderate complexity, suggesting that elastic
used in production are often limited by business needs, inversion can augment standard AVO analyses in
compute resources, and prior wavefield-extrapolation determining these properties. Helping the success of such
algorithm development. Hence, there may be increased inversions is the incorporation of at least some use of the
interest residuals for the tomographic operator that are Hessian to aid in parameter separation. For long-wavelength
created from gather techniques being used in RTM recovery, many of the methods discussed in the previous
production, such as the generation of Poynting vector sections extend naturally to the elastic regime.
gathers, and techniques that are used for trace alignment in
stack optimization. Acknowledgements
The author would like to thank Chevron Energy Technology
Company for permission to publish this work. Many thanks
to Ken Bube for a critical reading of the manuscript.
© 2017 SEG Page 5246

Figure 1. Conventional acoustic FWI. Figure 1a (left) shows an inversion to 6 hz overlaid on seismic image, Figure 1b (center) an inversion to 15
hz, and Figure 1c (right) a reflectivity image obtained from an inversion for impedance done to 20hz, analogous to Figure 1b. Courtesy Akela
Silverton, Nikhil Shah, and Vanessa Brown.
Figure 2. Depth slice result of 3D acoustic WEMVA based on RTM. Figure 1a (left) is the starting model, Figure 1b (center) is a standard ray-
based tomography result, Figure 1c (right) is the WEMVA result starting from the model in Figure 1a. The method used here employed DSO and
stack power optimization, as well as suitable amplitude normalization. Courtesy Guojian Shan and Yue Wang.
Figure 3. Tomographic image-domain FWI for salt. Figure 1a (left) shows the correct model, with a 5000m/s salt body, and 2000 m/s sediments.
The model dimensions are the same as Marmousi, so the salt is much wider than it appears. Figure 1b (second left) shows a limited bandwidth
standard RTM image without illumination compensation using a background model of 2000 m/s. The base salt cannot be seen; all events below
salt are intrasalt multiples. Figure 1c (third left) shows a subsurface offset gather from the center of the model. Even though the model is correct,
the strongest event is the top salt event, which is not well focused at all due to the contrast and bandwidth limits. Figure 1d (center) shows the result
of demigrating the stacked image and remigrating to a subsurface offset gather in the background model. The residual for the tomographic operator
was then obtained by spectral balancing and amplitude matching the gather in Figure 1c to that in Figure 1d and subtracting to give the gather in
Figure 1e (third right). This residual should be more representative of velocity error, with the strong top salt being significantly attenuated. Figure
1f (second right) shows a standard DSO WEMVA gradient for this data without any regularization, showing significant edge and intrasalt artifacts
from the multiples. Figure 1g (right) shows the gradient using the residual constructed in Figure 1e. Many of the WEMVA artifacts are now gone,
and the salt can be seen as its proper crescent shape, which is consistent with an update from a base salt event that is severely mispositioned.
Figure 4. 2D/3D elastic anisotropic/orthorhombic FWI for target sands below salt from hydrophone data. Figures 1a,b (left) show the results of a
2D inversion starting from a model that had no target sands below salt, but was otherwise correct. Figure 1a shows the recovery of vp, while Figure
1b shows the recovery of vs, with density being held fixed. The recovered model in 2D was very close to the true model. Figures 1c,d (left) show
the analogous 3D result. In this case the anisotropic coefficients were set to (𝜖1 , 𝜖2 , 𝛿1 , 𝛿2 , 𝛿3 , 𝛾1 , 𝛾2 ) = (. 09, .1, .04, .04, −.01,0.3,0.3) and were
multiplied by a spatially varying field that resembled the background sediment model. Recovery of shear and compressional velocity for the sand
under the rugose nose of salt was quite good in this case. Courtesy Peng Shen and Anusha Sekar.
© 2017 SEG Page 5247

EDITED REFERENCES
REFERENCES
Albertin, U., P. Shen, A. Sekar, T. Johnsen, C. Wu, K. Nihei, and K. Bube, 2016, 3D orthorhombic elastic
full-waveform inversion in the reflection domain from hydrophone data: 86th Annual
Alkalifah, T., M. Masmoudi, and J. Oh, 2016, A recipe for practical full-waveform inversion in
orthorhombic anisotropy: The Leading Edge, 1076–1083,
Biondi, B. and P. Sava, 1999, Wave-equation migration velocity analysis: 69th Annual International
Meeting, SEG, Expanded Abstracts, 1723–1726, http://dx.doi.org/10.1190/1.1820867.
Biondi, B. and A. Almomin, 2012, Tomographic full wavefrom inversion by combining full wavefrom
inversion with wave-equation migration velocity analysis: 82nd Annual International Meeting,
SEG, Expanded Abstracts, 1–5, https://doi.org/10.1190/segam2012-0275.1.
Fei, T., Y. Luo, F. Qin, and P. Kelamis, 2012, Full waveform inversion without low frequencies: A
synthetic study: 82nd Annual International Meeting, SEG, Expanded Abstracts, 1–5,
Fei, W., and P. Williamson, 2010, On the gradient artifacts in migration velocity analysis based on
differential semblance optimization: 80th Annual International Meeting, SEG, Expanded
Abstracts, 4071–4076, http://dx.doi.org/10.1190/1.3513710.
Kun, J., D. Sun, X. Cheng, and D. Vigh, 2015, Adjustive full waveform inversion: 85th Annual
Lou, Y., and G. Schuster, 1991, Wave-equation traveltime inversion: Geophysics, 56, 645–653,
https://doi.org/10.1190/1.1443081.
Ma, Y. and D. Hale, 2013, Wave-equation reflection traveltime inversion with dynamic warping and full-
waveform inversion: Geophysics, 78, R223–R233, http://dx.doi.org/10.1190/geo2013-0004.1.
Lazaratos, S., I. Chikichev, and K. Wang, 2011, Improving the convergence rate of full wavefield
inversion using spectral shaping: 81st Annual International Meeting, SEG, Expanded Abstracts,
2428–2432, http://dx.doi.org/10.1190/1.3627696.
Metevier, L., R. Brossier, Q, Merigot, E. Oudet, and J. Virieux, 2016, Increasing the robustness and
applicability of full waveform inversion: An optimal transport distance strategy: The Leading
Edge, 1060–1067, http://dx.doi.org/10.1190/tle35121060.1.
Mora, P., 1988, Elastic wave-field inversion of reflection and transmission data: Geophysics, 53, 750–
759, http://dx.doi.org/10.1190/1.1442510.
Mora, P., 1989, Inversion= migration+tomography: Geophysics, 54, 1575–1586,
http://dx.doi.org/10.1190/1.1442625.
Plessix R., P. Milcek, C. Corcoran, H. Kuehl, and K. Matson, 2012, Full waveform inversion with a
pseudotime approach: 74th Annual International Conference and Exhibition, EAGE, Extended
Abstracts, W012, http://dx.doi.org/10.3997/2214-4609.20148713.
Routh, P., R. Neelamani, R. Lu, S. Lazaratos, H. Braaksma, S. Hughes, R. Saltzer, J. Stewart, K. Naidu,
H. Averill, V. Gottmukkula, P. Homonko, J. Reilly, and D. Leslie, 2017, Impact of high-
© 2017 SEG Page 5248

resolution FWI in the Western Black Sea, Revealing overburden and reservoir complexity: The
Leading Edge, 36, 60–66, http://dx.doi.org/10.1190/tle36010060.1.
Shen, P., W. Symes, and C. Stolk, 2003, Differential semblance velocity analysis by wave-equation
migration title: 73rd Annual International Meeting, SEG, Expanded Abstracts, 2132–2135,
https://doi.org/10.1190/1.1817759.
Shen, P. and W. Symes, 2015, Horizontal contraction in the image domain for velocity inversion:
Geophysics, 80, R95–R110, http://dx.doi.org/10.1190/geo2014-0261.1.
Sirgue, L., O. Barkved, J. Dellinger, J. Etgen, U. Albertin, and J. Kommendal, 2010, Full-waveform
inversion: the next leap forward in imaging at Valhall: First Break, 28, 65–70.
Sun, D., J. Kun, X. Cheng, and D. Vigh, 2016, Reflection-based waveform inversion title: 86th Annual
Tang, Y., S. Lee, A. Baumstein, and D. Hinkley, 2013, Tomographically enhanced full waveform
inversion title: 83rd Annual International Meeting, SEG, Expanded Abstracts, 1037–1041,
Warner, M. and L. Guasch, Adaptive waveform inversion: Theory: Geophysics, 81, 429–445,
Wu, R. J. Luo, and B. Wu, 2014, Seismic envelope inversion and modulation signal model: Geophysics,
79, WA13–WA24, http://dx.doi.org/10.1190/geo2013-0294.1.
Xu, S., D. Wang, F. Chen, and Y. Zhang, 2012, Full waveform inversion for reflected seismic data: 74th
Annual International Conference and Exhibition, EAGE, Extended Abstracts,
http://dx.doi.org/10.3997/2214-4609.20148725.
Zhang, S., and G. Schuster, 2013, Image-domain full waveform inversion: 83rd Annual International
Meeting, SEG, Expanded Abstracts, 861–865, http://dx.doi.org/10.1190/segam2013-1238.1.
© 2017 SEG Page 5249

Automatic Velocity Model Building with Complex Salt:
Can Computers Finally Do an Interpreter’s Job?
Scott Michell*, Xukai Shen, Andrew Brenders, Joe Dellinger, Imtiaz Ahmed, and Kang Fu
BP America Inc., Houston, TX
Summary into the 50s, with alternative salt scenarios often yielding
seismic images of nearly equivalent quality. The limitations
The requirements for a good seismic image are simply of Kirchhoff migration in ray-tracing through a complex
stated: appropriate data, suitable algorithms, and a good overburden proved a particular problem when used for
velocity model. Some might add a “friendly earth” to the list, building salt bodies, and the computation time required for
since it is clear there are some areas where we have not yet pre-stack depth migration was still a significant roadblock.
discovered how to compute a good image, most notably
beneath rugose, highly variable media. Over the past two As we moved into the early 2000's and compute capability
decades, our seismic data and imaging algorithms have continued to increase, wave theoretic imaging algorithms
greatly improved. Increasing data density and wide azimuth such as common-azimuth wave-equation migration became
acquisition, combined with improved demultiple techniques cost efficient enough for us to use in high-end processing
and anisotropic wave-theoretic migration have produced projects (e.g., Biondi and Palacharla, 1996). Additionally,
significant uplifts in our ability to image beneath complex the advancement of demultiple techniques such as surface
salt. Velocity-model building has lagged behind: building related multiple elimination (SRME) from 2-D to 3-D
models with complicated salt bodies continues to be a labor- enabled a step-change in our ability to remove multiples - a
intensive, hit-or-miss interpretation exercise that often major roadblock to unambiguous interpretation of many
produces unreliable results. The promise of Full-Waveform deepwater prospects. These technologies allowed us to
Inversion (FWI) - a fully-automatic, data driven, model- significantly improve our seismic images, but velocity-
building process - has proved elusive, notwithstanding its model building remained time-consuming and problematic.
significant impact in non-salt settings. Recent experience Recognizing that NATS data were proving to be insufficient
with the Atlantis field in the Gulf of Mexico has shown that to image our most challenging subsalt reservoirs, BP
automatic velocity-model building with FWI is also possible embarked on developing and deploying two new acquisition
for salt. When applied to ocean-bottom seismic data technologies: Wide Azimuth Towed Streamer (WATS) and
acquired with wider offsets and lower frequencies than deepwater, wide azimuth Ocean Bottom Nodes (OBN).
previously recorded, FWI produced a velocity model that These new methods dramatically improved the seismic
dramatically improved the seismic image at Atlantis. images at Mad Dog and Atlantis in the 2005-2006
timeframe.
Introduction
Industry’s difficulty with building accurate velocity models
Industry's interest in deep water sub-salt imaging began in was not limited to deepwater, sub-salt prospects in the Gulf
the mid-1990's. At the time, the state of the art in seismic of Mexico. Challenges in seismic imaging below non-salt
acquisition was 3-D, Narrow Azimuth Towed Streamer complex overburden are widespread. Shallow gas at
(NATS), and significant work was still ongoing to move Valhall, mud volcanoes in the Caspian Sea, and karsts in
from post- to pre-stack migration, and from time- to depth- Indonesia are a few examples. BP tested various automatic
imaging. Kirchhoff depth migration and Radon demultiple velocity-model-building workflows in all these areas. One of
were the state-of-the-art technology. Reflection-based, the promising approaches identified was full-waveform
travel-time tomography had arrived on the scene, removing inversion (FWI).
the labor intensive and error prone layer-stripping methods
of building velocity models. However, using tomography for The Promise of Full-Waveform Inversion
salt model definition proved problematic, leading industry to
adopt an iterative, hybrid approach of using tomography to Developed in the early 1980’s by Tarantola and others, FWI
update the velocity model in the sediments, then migrating, updates velocity models iteratively by matching forward
interpreting the salt body, flooding with salt velocities, modeled data with field data (e.g., Tarantola, 1984). With a
picking the shape of the salt body, inserting this body into theoretical resolution limited only by the bandwidth of the
the surrounding sediment velocities, and repeating. This data used, automatic velocity-model building by FWI has
remains the standard approach for velocity-model building been a promise since its initial conception. Practical
in areas with salt. As the complexity of the salt body and the applications of FWI remained limited until the early 2000’s,
number of inclusions increases, the number of iterations since the method requires the computational expense of
required to converge to a suitable migrated image rises, often modeling with a two-way wave equation. Industry interest
© 2017 SEG Page 5250

Automatic Velocity Model Building: Computers vs Humans
was piqued in 2004 by Pratt and Brenders’ results with a shown in Figure 2, which shows RTM images using the most
blind synthetic dataset created for an EAGE workshop recent OBN data, but migrated through velocity models
(Billette and Brandsberg-Dahl, 2005). By 2006, Sirgue et al. before and after our most recent FWI (see Shen et al., 2017).
(2008) had developed an efficient workflow using time- The effect on the image is significant, and has had a direct
domain propagators, and FWI finally became feasible for 3- and immediate economic impact on the development of the
D applications at industrial scale. Atlantis Field. How was this possible?
Commercial success was first achieved with wide azimuth
Essential Ingredients for FWI
OBC data over the Valhall field in the North Sea, proving
the efficacy of FWI for velocity-model building given the The 2015 Atlantis OBN survey was acquired with a 4330 cu.
right conditions. Since 2006, FWI has proved to be a in. airgun array towed at a 15 m depth, the same array design
significant advancement in shallow water sedimentary used in 2005 (Lewis et al., 2016). Figure 3 shows airgun
basins, fundamentally changing how BP images below energy spectra calculated for a line of nodes across the
shallow gas (Sirgue et al., 2010). We have continued to crossline axis of the array, so that we show shots spanning
widely apply FWI in offshore basins such as Trinidad and the full time span of the survey. For each node, we calculate
Azerbaijan. Industry has also continued to investigate a far-field source spectrum in SEG dB units using an
applications of FWI to building velocity models in areas of overhead airgun shot. We can see that above ~5 Hz the far-
complex salt, but with only limited success. field output of the airguns in 2005 and 2015 was indeed very
similar, as expected for a 4D reshoot.
Successful full-waveform inversion requires three
fundamental ingredients: appropriate data, a good starting Below 5 Hz, the downward slope of the 2015 airgun spectra
model, and an appropriate workflow / algorithm. Due to the is noticeably shallower. There is also considerably more
non-linearity of the problem, the need for low-frequency variability in amplitude between shots in 2005 when
data is well known, a need which can be mitigated by having compared with the 2015 acquisition, indicating a more stable
an extremely good starting model, less so by recording wide- source signature in the later acquisition. Below ~1.5 Hz, the
angle data. In 2012, Chevron produced a synthetic, elastic curves slope upwards with decreasing frequency. This is
dataset over a salt model without low frequencies in the data. background noise, which rapidly increases below ~4 Hz with
Results from multiple contractors and researchers decreasing frequencies. The bottom of the valley shows
demonstrated that FWI could not update the salt (e.g., where the signal-to-noise ratio (S/N) is about unity, as the
Williamson et al., 2016), even with long-offset data. dominant energy source transitions from the airguns above 4
Numerous variations in the FWI algorithm / workflow have Hz to the background noise below 1.5 Hz. In 2005 (black
been tried, but these fundamental requirements have arrow), this crossover was at ~2.5 Hz. In 2015 (red arrow),
remained. it was ~1.6 Hz, about 35% lower. We believe that the
improved signal at low frequencies is primarily due to more
Automatic Salt Model Building Is Possible careful airgun array QC during acquisition, while the
reduced noise is due to improved node electronics.
In 2015, BP executed an OBN survey over Atlantis, intended
as a repeat to the 2005-2006 survey for 4-D processing (van There is a trade-off between the value of low-frequencies
Gestel et al., 2015). In addition to 4-D imaging, the survey and long-offsets (wide angles) in FWI, so what is imporant
was designed to improve the image beneath the salt by is an understanding of the S/N as a function of offset. Figure
acquiring a dataset with the ingredients preferred for model 4 shows the phase of 2.25 Hz data calculated from a single
updating with FWI. Data were recorded to about double the ocean-bottom node, from both 2005 and 2015. These plots
offset required to effect an exact 4-D repeat. This was demonstrate the clear improvement in S/N at long offsets for
possible because of improved node battery life and node 2.25 Hz in 2015 versus 2005. Also note the much greater
numbers sufficient for a single patch acquisition. Great care offsets recorded in 2015 versus 2005.
was taken to keep the airgun arrays well calibrated, possible
because the longer battery life also allowed tighter source In exploration settings, the confidence in our starting
specifications. FWI applied to this dataset made substantial velocity model is likely to be much lower than at Atlantis.
revisions to the salt in the velocity model in a way never seen This implies that to achieve a similar level of success, we
before. will need to record even lower frequency / wider offset data.
How much more will we need? This remains an open
Figure 1 shows the evolution of an inline section of the question, but the capability to produce such data already
velocity model at Atlantis, from (a) 2004, (b) 2013, and (c) exists (Dellinger et al., 2016). Figure 5 shows calibrated
2016. With the model rebuild from 2013 as the starting synthetic modeling of what should be possible with a
model, FWI using the 2015 OBN dataset was able to build Wolfspar® source at 1.6 Hz, compared with what we
the model in Figure 1(c) automatically, without human recorded at Atlantis in 2015. Modeling suggests this data
interpretation of the salt body. The impact of this model is
© 2017 SEG Page 5251

quality is in the range of what we need for FWI to work with

exploration-quality starting models.
Challenges in economically acquiring the long offsets

illustrated by Figure 5 remain. Due to the Nyquist sampling
requirements of FWI, we are lead to a “sparse 3-D” style of
acquisition, with crossline spacing much wider than inline
spacing for both shots and receivers, and the shot inline
direction orthogonal to the receiver inline direction.
Assuming WAZ data already exist over the prospect for
imaging, and we only need the wide offsets for velocity-
model building, the crossline axis can be very coarse. Model
studies suggest that crossline spacings of multiple
kilometers may be sufficient.
Figure 1: A section through the 3-D velocity model at Figure 2: Migrated images of the 2015 Atlantis OBN dataset,
Atlantis, showing evolution of the model through several using the velocity models shown in Figure 1(b) and (c),
interpretation / processing cycles: (a) the isotropic model respectively. The velocity model from 2016, including a salt
with a homogeneous salt body, circa 2004, (b) the body automatically updated by FWI, dramatically improved
anisotropic velocity model from 2013, still manually built, the subsalt image. Both images were calculated using an
but with considerable guidance from FWI, and (c) the identical anisotropic reverse time migration algorithm.
anisotropic velocity model built automatically by FWI,
starting from the 2013 model in (b).
© 2017 SEG Page 5252

Figure 3: Representative near-offset frequency spectra of the

2005 Atlantis OBN dataset (black) and its 4D repeat in 2015
(red). Airgun arrays were identical in both size (4330 cu.
in.) and design, and both produced similar signal strengths
above 5 Hz. However, below 5 Hz the 2015 acquisition
recorded both higher signal and lower noise. After Dellinger
(2016).
Conclusions
BP has demonstrated at Atlantis we can attain ~2 Hz data at
sufficiently long offsets to enable FWI to meaningfully
update the velocity model. The observations are consistent
with the synthetic study outlined by Dellinger et al. (2017)
and we have since run this workflow on other datasets to
demonstrate its robustness. However, our model study Figure 4: Common-receiver-gathers (CRG) showing the
suggests that given a starting velocity model that is phase of 2.25 Hz data from the 2005 (top) and 2015 (bottom)
sufficiently wrong, the frequencies we typically record will Atlantis OBN datasets, plotted at the same map scale. The
not be low enough. This is why BP has been investing in a 2015 acquisition recorded data with maximum offsets
new type of Low Frequency Seismic source, Wolfspar®. appoximately double those recorded during the 2005 survey,
and demonstrate improved S/N at low frequencies. The
The proven ability of FWI to automatically update a velocity yellow dashed box compares the extent of a 2005 CRG to
model in a complex salt setting is exciting. It holds the one recorded in 2015.
promise of drastically reducing cycle time while also
improving the image. To realize this vision more generally,
we will need to acquire longer-offset OBN data with a lower-
frequency source. This demands much more efficient
acquisition of those surveys in order to move from bespoke
production surveys to large scale exploration surveys.
Putting it all together, we have a vision of a possible future:
salt models built automatically by FWI, using low-
frequencies from non-impulsive sources tailored to the
frequencies required for FWI, with long offsets recorded in
a “Sparse 3-D” acquisition. With these data, FWI may
finally be able to build robust velocity models suitable for
sub-salt imaging, without the need for iterative, human
interpretation. Computers may be able to do an interpreter’s
job! Figure 5: A simulated CRG phase ring plot from BP’s
Wolfspar® prototype low-frequency seismic source at 1.6 Hz.
Acknowledgements Field recordings of the signal and noise were used to calibrate
We thank BP and the partner of the Atlantis Field (BHP the simulation to produce realistic signal-to-noise ratios as a
Billiton) for permission to publish this paper. function of offset. The red inset figure shows the phase of 1.6
Hz field data from the 2015 Atlantis acquisition, recorded
using conventional airgun sources, from the same CRG as in
Figure 4.
© 2017 SEG Page 5253

EDITED REFERENCES
REFERENCES
Billette, F. J., and S. Brandsberg-Dahl, 2005, The 2004 BP velocity benchmark: 67th Annual
International Conference and Exhibition, EAGE, Extended Abstracts, B035,
http://earthdoc.eage.org/publication/publicationdetails/?publication=1404.
Biondi, B., and G. Palacharla, 1996, 3-D prestack migration of common-azimuth data: Geophysics, 61,
1822–1832, http://dx.doi.org/10.1190/1.1444098.
Dellinger, J., 2016, Challenges to extending the usable seismic bandwidth at the seafloor in the deepwater
Gulf of Mexico: 86th Annual International Meeting, SEG, Expanded Abstracts, 66–70,
Dellinger, J., A. J. Brenders, J. R. Sandschaper, C. Regone, J. Etgen, I. Ahmed, and K. J. Lee, 2017, The
Garden Banks model experience: The Leading Edge, 36, 64–71,
Dellinger, J., A. Ross, D. Meaux, A. Brenders, G. Gesoff, J. EtÂgen, J. Naranjo, G. Openshaw, and M.
Harper, 2016, WolfÂspar®, an “FWI-friendly” ultralow-frequency marine seismic source: 86th
Lewis, B., C. Brooks, M. Pfister, S. Michell, and G. Astvatsaturov, 2016, Efficient acquisition of
deepwater node surveys: 86th Annual International Meeting, SEG, Expanded Abstracts, 92–96,
Shen, X., I. Ahmed, A. Brenders, J. Dellinger, J. Etgen, and S. Michell, 2017, Full-waveform inversion:
the next leap forward in subsalt imaging: The Leading Edge, in preparation.
Sirgue, L., O. I. Barkved, J. Dellinger, J. Etgen, U. Albertin, and J. H. Kommedal, 2010, Full waveform
inversion: the next leap forward in imaging at Valhall: First Break, 28, 65–70,
http://dx.doi.org/10.3997/1365-2397.2010012.
Sirgue, L., J. T. Etgen, and U. Albertin, 2008, 3D frequency domain waveform inversion using time
domain finite difference methods: 70th Annual International Conference and Exhibition, EAGE,
Extended Abstracts, http://dx.doi.org/10.3997/2214-4609.20147683.
Tarantola, A., 1984, Inversion of seismic reflection data in the acoustic approximation: Geophysics, 49,
no. 8, 1259–1266, http://dx.doi.org/10.1190/1.1441754.
Van Gestel, J.-P., E. L’Heureux, J. R. Sandschaper, P.-O. Ariston, N. D. Bassett, and S. Dadi, 2015,
Atlantis “Beyond 4D” ocean bottom nodes acquisition design: 85th Annual International
Meeting, SEG, Expanded Abstracts, 125–129, http://dx.doi.org/10.1190/segam2015-5847522.1.
Williamson, P., C. Rivera, and K. Mansoor, 2016, Acoustic FWI applied to subsalt imaging: An
illuminating case study from offshore Angola: 86th Annual International Meeting, SEG,
Expanded Abstracts, 4896–4900, http://dx.doi.org/10.1190/segam2016-13964890.1.
© 2017 SEG Page 5254

Achieving business value using high-end borehole geophysics: Are we there yet?
Brian E Hornby, Halliburton
Summary initial results most often do not represent actual achieved

business value. Rather, they more likely represent
With recent advances in high-end borehole geophysics, it something of academic interest that requires more
might be prudent to question what is actually necessary to development to achieve real value.
achieve business value and whether such practices are
currently in place. Here, “high-end” borehole geophysics This can result in premature commercialization by vendors
essentially goes beyond the standard time-to-depth and increased expectations for operators. A case in point is
measurement with major service providers, such as three- 3D VSP imaging.
dimensional (3D) imaging, time-lapse reservoir monitoring,
passive (micro-seismic) monitoring, and surface seismic 3D VSP imaging has been used since the 1990s and has
(anisotropic) velocity model calibration. Instead, this study been a development process. Early examples showed
focuses primarily on marine 3D vertical seismic profiling promise, but clear business value was difficult to determine
(VSP) imaging and reservoir monitoring applications, with because of poor image quality, largely resulting from
consideration to other applications. inadequate tool and imaging solutions. Historic barriers to
success include overoptimistic feasibility modeling,
normally produced by a basic illumination process that
Introduction poorly predicted image quality for a given survey design.
Another issue is centered around downhole tool
The majority of VSP surveys are for time-to-depth curves, technology—fundamentally, VSP imaging is challenged
primarily for seismic well tie. Although seismic interpreters because of inherent poor angular coverage caused by
require a simple answer, this basic use of VSP might not be having all receivers in a well. Clearly, it would be ideal to
simple, particularly in wells penetrating or in close fully instrument the well, which was not possible with early
proximity to complex geology (e.g., a salt body). Careful technology because of array technology limitations. In
modeling and survey design are needed for optimal results. addition, 3D velocity models were also an issue. With
these, the only choice, unless being truly in a layer cake
This need for careful modeling and survey design is even with usable well data, is the 3D surface seismic model. An
more important for high-end borehole seismic applications. accurate model was necessary that represented the complex
Considering, in particular, 3D VSP imaging and time-lapse and anisotropic properties of the earth. Initially, an
reservoir monitoring; historic barriers for delivering isotropic model was possible that, in reality, represented the
business value have stemmed from limitations in both the best velocity to focus the two-way arrivals generated and
algorithm/processing side and the downhole and surface received at the surface, but it did not represent an accurate
hardware. Critical enabling technology includes advances velocity in any dimension, having the property of a
in high-performance computing that enable employing distorted velocity resulting from a compensation for the
more complex algorithms and larger downhole arrays that subsurface anisotropy. Because VSP data do not represent
can deliver the necessary imaging coverage and resolution ray-paths to and from the surface, the distorted isotropic
to be useful for seismic interpretation. For downhole model that best matched the surface seismic data was
technology, a new technology is distributed acoustic wrong for the VSP geometry, which is particularly
sensing (DAS), which can achieve large downhole arrays at detrimental because the goal normally is to achieve a
almost no cost using fiber already in wells leading to higher-resolution product rather than a distorted view
pressure/temperature (P/T) gauges or by inexpensively caused by the incorrect model.
adding fiber to a cable pack for a new well. Of primary
concern is determining the trade-off between DAS and Recent Advances with 3D VSP Imaging
dedicated receiver arrays and, more importantly, what is
required to achieve success for a given business challenge. The current state of technology warrants discussion. For
hardware, the focus has been on development of longer
arrays specifically intended for borehole seismic imaging.
Barriers to Achieving Business Value For wireline deployed systems, array sizes for 3D VSP
acquisition deployed in the deepwater Gulf of Mexico have
New technology advances can create excitement in the increased from 600 ft in 2001 (Ray et al., 2003) to 4,000 ft
industry, which might lead to a bandwagon mentality. This (Graves et al., 2008) to more than 6,000 ft presently (Li et
excitement can be generated oftentimes by initial intriguing al., 2016).
results presented at public conferences; however, these
© 2017 SEG Page 5255

While significant progress has been achieved for downhole Surface source technology also has evolved. The greatest
array technology, it might be necessary to ask whether advance relevant to 3D VSP imaging is simultaneous multi-
these array sizes are sufficiently large or if more progress is source acquisition (Beasley et al., 1998; Stefani et al.,
needed. Array size depends on the particular imaging 2007; Abma et al., 2015). Deepwater offshore rigs can cost
challenge. In Jiang et al. (2012), modeling showed the best upwards of USD 1 million per day, so simultaneous
results using an 8,000 ft array, but acceptable results were shooting with three source vessels can reduce the
produced for a 6,600 ft array. The technical limit on array acquisition time by almost a factor of three and allow for
size will vary with the issue, but clearly here, more is large and dense shot patterns that can greatly increase the
better. subsurface imaging coverage and image quality.
Concerning velocity models, marine surface seismic Multi-well Imaging and Time-lapse Reservoir
imaging has moved in recent years to anisotropic (TTI) Monitoring Using Permanent Borehole Seismic
velocity models. This significantly helped image quality for
3D VSP imaging, as eliminating the distorted isotropic Although time-lapse monitoring can be performed using
model and providing results more representative of the multiple 3D VSP surveys by leaving access to the well
actual earth properties resulted in clearer images and, available for a return deployment, this is only possible for
provided the model is reasonable, a good association with limited cases where a dedicated monitor well is in place
the VSP imaging to the surface seismic result rather than and is not possible for typical wells that are going to be
having to stretch and squeeze the imaging result, as was completed for production or injection. Therefore, focus is
common with isotropic models. on the permanent deployment of downhole sensors, which
also allows instrumenting multiple wells across a reservoir.
Another advance representing a key enabling technology is With multiple wells instrumented, the goal expands to
high-performance computing (HPC). Since the early 1990s, include the possibility of large-scale, high-resolution
the speed of HPC systems has increased by approximately reservoir imaging. Here, the ultimate success would be to
an order of magnitude every few years to the current peak replace, for development, seismic imaging and four-
of more than 100 Petaflops (Strohmaier et al., 2017). This dimensional (4D) surface or seabed seismic with multiple
significant increase in speed has resulted in many changes wells instrumented with arrays of borehole seismic
to the seismic imaging industry, enabling the following: receivers. In addition to enabling larger imaging patches,
the multiple well arrays should also improve imaging
1. Large-scale and efficient 3D finite-difference method quality, with better angular coverage and correspondingly
(FDM) modeling: this is key to the up-front feasibility fewer artifacts than single well solutions.
and survey design process and has proven
instrumental in defining both the imaging goal and Downhole hardware is key. Satisfying this vision requires
survey requirements before initiating a survey (Jiang a safe, reliable, and cost-effective method to instrument the
et al., 2012; Rollins et al., 2015) wells with the necessary array size and P/T parameters
2. Practical use of advanced imaging algorithms: two- while providing low risk to the well completion process.
way wave equation imaging reverse-time migration “Cost effective” does not necessarily mean “low cost.” For
(RTM) had algorithms available in the early 2000s; example, if the result will affect placement of half billion
however, only in recent years could the dollar wells in a prolific oil field, then the cost might be
computationally expensive algorithm be practically relatively high but still acceptable. Conversely, for a land
and routinely applied for realistic velocity models well in an unconventional shale play, spending one million
(e.g., 20  20  10 km) and realistic frequencies (50 dollars on a downhole kit would normally not be
Hz or higher is now possible). RTM has proven to be acceptable.
important for imaging complex subsalt reservoirs and
is considered to be a major advantage for 3D VSP Hardware for permanent borehole seismic involves either
imaging, where more accurate imaging coupled with instrumentation of traditional three-component (3C) or
the flexibility to image with both primaries and free- four-component (4C) sensors, or using DAS.
surface multiples has been key. Beyond RTM
imaging, developing imaging solutions includes even For traditional sensors, tubing-deployed permanent fiber-
more expensive algorithms, such as an iterative least- optic borehole seismic sensors were installed in two wells
squares solution. Additionally, full-waveform located in the Norwegian North Sea and in the deepwater
inversion (FWI) is routinely applied to surface seismic Gulf of Mexico (Hornby et al., 2012). The North Sea well
data; it has had limited application but holds promise had five levels spaced 13 m apart, and the deepwater Gulf
for 3D VSP application. of Mexico well had 12 levels spaced 23 m apart. Results
from the 3D VSP data acquired in these wells showed
© 2017 SEG Page 5256

images that were higher-resolution and consistent with The Next Step: Multi-well Imaging and Reservoir
OBN seismic images; however, the authors concluded that Monitoring with DAS
to achieve a significant business impact, much larger

borehole arrays, ideally instrumenting most of the borehole, In good conditions, comparable or perhaps even better
are needed to produce the required image coverage and imaging with DAS is possible. The actual goal is two-fold:
quality. reservoir scale imaging with multiple instrumented wells
and efficient and low-cost permanent 4D reservoir
Another recent project that is underway is attempting to monitoring (Lopez et al., 2012; Chalenski et al., 2016). As
configure a much larger tubing-deployed borehole seismic of this writing, the demonstration of permanent reservoir
array using a hybrid design consisting of traditional monitoring using DAS in a deepwater setting has not been
electrical 3C sensors and downhole electronics, coupled shown, though monitor data were acquired at the same
with fiber-optic transmission of the large data volumes to Mars field, where good-quality imaging was previously
the surface (Gerea et al., 2016). This project is aimed at demonstrated (Chalenski et al., 2016).
both imaging uplift and 4D reservoir monitoring and will
use 115 shuttles with a 1425 m long array. Remaining Challenges for DAS
DAS: a Major Breakthrough in Downhole Array 1. Although DAS surpasses the “fully instrument the
Technology? well” barrier and is certainly low-cost, challenges
remain. Signal-to-noise (S/N) is the first challenge that
DAS for imaging and permanent borehole monitoring has DAS has historically experienced, with comparison of
created much interest. DAS simply uses the fiber cable that DAS to traditional geophone response showing a 20
normally would be the lead-in cable to downhole sensors dB reduction (e.g., Li et al., 2013). Presently, DAS
(P/T or seismic), and the seismic system is a box on the service providers are aware of this challenge and have
surface platform (Mateeva et al., 2012). been adding improvements to the DAS box as well as
working with fiber providers to obtain fiber dedicated
Therefore, in effect, a passive fiber strapped to a production to imaging, as opposed to data sets acquired up to the
tubing could be used to fully instrument a well; even if date of this writing using available fiber intended as
fiber sensors are not planned for use in a well, a fiber can lead-in cables for downhole sensors and not for
be embedded in the standard cabling flat-pack that includes imaging. Therefore, for new well delivery, the
the electrical connections to the standard downhole P/T combination of both DAS technology upgrades with
gauges. Potentially, any well with a downhole fiber-optic improved, more sensitive fiber is hoped to remove
P/T gauge could be fully instrumented with large seismic most or all of the difference between DAS and
arrays. standard geophones in terms of S/N.
2. Depth registration is a known challenge for DAS
Imaging with DAS (Mateeva et al., 2014; Verliac et al., 2015). Verliac et
al. (2015) compare tests of four DAS operators and
Recent advances in DAS show promise for large-scale show a depth registration difference of approximately
imaging in a marine setting. The first 3D DAS survey 17 to 30 m between the different operators. Additional
acquired by an operator in a deepwater well (Mateeva et al., effort was made to develop procedures for depth
2013) is considered a landmark achievement. Subsequent registration by Ellmauthaler et al. (2016), who
processing of those data showed imaging results at least analyzed depth discrepancies between a standard VSP
comparable to 3D ocean-bottom node (OBN) seismic survey and a DAS survey in the same well. It was
images (Wu et al., 2015; Zhan et al., 2015). Another marine concluded, for the particular case studied, that DAS
survey in Trinidad also showed comparable images for VSP depths can be determined with reasonable
OBN seismic and DAS over the aperture of the DAS accuracy using two calibration points along the well.
survey (Zhan et al., 2015). An additional marine survey
was recently acquired at Valhall, and data were acquired Conclusions: Are We There Yet?
simultaneously during the life of field seismic (LoFS)
ocean-bottom seismic (OBS) survey (Jiang et al., 2016). In 1. Wireline deployed 3D VSP imaging. It was
that survey, encouraging results were observed for imaging established that array size is an important component
with DAS while the instrumented well was producing, as for success. For 3D VSP, recent advances in array size
comparable images to OBN images were observed at exceed 6,000 ft. One modeling effort shows a 6,600 ft
intermediate depths; however, DAS failed to produce array tool planned for the survey would deliver
usable images at reservoir depths. required results as dictated by the seismic interpreters;
however, it was noted that a longer 8,000 ft survey
© 2017 SEG Page 5257

would deliver better results. Therefore, it can be

concluded that the current >6,000 ft array size
technology can satisfy much of the market, but arrays

in the 8,000 to 10,000 ft range would be optimal for
the deepwater market. Again, careful modeling is
needed to establish what is required for any survey.
2. Multi-well imaging and time-lapse reservoir
monitoring using permanent borehole seismic.
Dedicated receiver imaging and 4D technology are
still in early stages, and initial field trials are needed to
understand any remaining challenges. DAS has shown
major advances but has remaining technical
challenges, as noted previously.
3. Other challenges for permanent sensors—
infrastructure. One issue not discussed here is
whether permanent sensor deployment in complex
subsea completion wells has any current barriers. For
DAS and at least some dedicated sensor solutions,
deployment of the fiber-optic cable for DAS and optic
pass-through for fiber are needed. Various projects
have studied this issue; however, to date, no
commercial solution for plugging the fiber cable
through an available port in a subsea tree is available.
4. Remaining technical challenges. As of this writing,
the final endorsement for 3D VSP is lacking.
Operating companies still need to demonstrate that 3D
VSP results add tangible business value, such as in the
placement of a producer well to accompany an
injection well in which the survey was acquired. The
same issue exists for imaging and reservoir monitoring
with permanent sensors. For example, we can claim
success when an operating company shows that time-
lapse monitoring has accurately identified bypassed oil
and a well is at least targeted to drain that
compartment.
© 2017 SEG Page 5258

EDITED REFERENCES
REFERENCES
Abma, R., D. Howe, M. Foster, I. Ahmed, M. Tanis, Q. Zhang, A. Arogunmati, and G. Alexander, 2015,
Independent simultaneous source acquisition and processing: Geophysics, 80, no. 6, WD37–
WD44, https://doi.org/10.1190/geo2015-0078.1.
Beasley, C. J., R. E. Chambers, and Z. Jiang, 1998, A new look at simultaneous sources: 68th Annual
International Meeting, SEG, Expanded Abstracts, 133–135, https://doi.org/10.1190/1.1820149.
Chalenski, D., M. Tatanova, Y. Du, A. Mateeva, J. Lopez, and H. Potters, 2016, Climbing the staircase of
ultralow-cost 4D monitoring of deepwater fields using DAS-VSP: 86th Annual International
Meeting, SEG, Expanded Abstracts, 5441–5445, https://doi.org/10.1190/segam2016-13681284.1.
Ellmauthaler, A., M. Willis, D. Barfoot, X. Wu, C. Erdemir, O. Barrios-Lopez, D. Quinn, and S. Shaw,
2016, Depth calibration for DAS VSP: Lessons learned from two field trials: 86th Annual
International Meeting, SEG, Expanded Abstracts, 632–636, https://doi.org/10.1190/segam2016-
13951487.1.
Gerea, C., P.-L. Pichon, G. Geddes, M. Verliac, and V. Lesnikov, 2016, Proficient subsalt monitoring
with 3D well seismic, deep-offshore West Africa: 86th Annual International Meeting, SEG,
Expanded Abstracts, 688–692, https://doi.org/10.1190/segam2016-13869509.1.
Graves, J., S. Checkles, J. Leveille, A. Campbell, S. Leaney, and C. P. Deri, 2008, 3D VSP acquisition
and 3C processing on a deep subsalt prospect in the Gulf of Mexico: 78th Annual International
Meeting, SEG, Expanded Abstracts, 3340–3344, https://doi.org/10.1190/1.3064038.
Hornby, B. E., O. Barkved, H. Sugianto, and J. Nahm, 2012, 4D monitoring with permanent borehole
seismic sensors: progress, challenges, and the road ahead: 82nd Annual International Meeting,
SEG, Expanded Abstracts, 1–5, https://doi.org/10.1190/segam2012-1519.1.
Jiang, T., G. Zhan, T. Hance, S. Sugianto, S. Soulas, and E. Kjos, 2016, Valhall dual-well 3D DAS VSP
field trial and imaging for active wells: 86th Annual International Meeting, SEG, Expanded
Jiang, Z., B. Hornby, and D. Stauber, 2012, A model study investigating enhanced subsurface imaging
and fault detection using multiple-well 3D VSP acquisition, migration, and coherency analysis:
82nd Annual International Meeting, SEG, Expanded Abstracts, 1–5,
Li, Q., A. Reitz, T. Jia, K. Hartman, F. Rollins, W. Slopey, G. Deren, S. Michell, J. Naranjo, R. Abma,
and M. Mahnke, 2016, Two record-setting subsalt 3D VSP acquisitions in the Gulf of Mexico
2015: Experience and learnings: 86th Annual International Meeting, SEG, Expanded Abstracts,
Li, Q., B. Hornby, and J. Konkler, 2013, A permanent borehole fiber-optic distributed acoustic sensing
experiment: 83rd Annual International Meeting, SEG, Expanded Abstracts, 5057–5061,
Lopez, J. L., D. Kiyashchenko, A. Mateeva, J. Mestayer, W. Berlang, P. Wills, B. Cox, G. Rocco, R.
Adawi, D. Hill, P. Martin, and J. Roy, 2012, 3D VSP for affordable permanent reservoir
monitoring onshore and offshore: 82nd Annual International Meeting, SEG, Expanded Abstracts,
Mateeva, A., J. Lopez, H. Potters, J. Mestayer, B. Cox, D. Kiyashchenko, P. Wills, S. Grandi, K.
Hornman, B. Kuvshinov, W. Berlang, Z. Yang, and R. Detomo, 2014, Distributed acoustic
© 2017 SEG Page 5259

sensing for reservoir monitoring with vertical seismic profiling: Geophysical Prospecting, 62,
679–692, https://doi.org/10.1111/1365-2478.12116.
Mateeva, A., J. Mestayer, B. Cox, D. Kiyashchenko, P. Wills, J. Lopez, S. Grandi, K. Hornman, P.
Lumens, A. Franzen, D. Hill, and J. Roy, 2012, Advances in distributed acoustic sensing (DAS)
for VSP: 82nd Annual International Meeting, SEG, Expanded Abstracts, 1–5,
Mateeva, A., J. Mestayer, Z. Yang, J. Lopez, P. Wills, J. Roy, and T. Bown, 2013, Dual-well 3D VSP in
deepwater made possible by DAS: 83rd Annual International Meeting, SEG, Expanded Abstracts,
Ray, A., B. Hornby, and J. P. Van Gestel, 2003, Largest 3D VSP in the deep water of the Gulf of Mexico
to provide improved imaging in the thunder horse south field: 73rd Annual International Meeting,
SEG, Expanded Abstracts, 422–425, https://doi.org/10.1190/1.1817939.
Rollins, F., J. R. Sandschaper, Q. Li, F. Ye, and S. Chakraborty, 2015, Calibrating 3D VSP image area
with finite difference modeling – A case study in Gulf of Mexico: 85th Annual International
Meeting, SEG, Expanded Abstracts, 3517–3521, https://doi.org/10.1190/segam2015-5873629.1.
Stefani, J., G. Hampson, and E. F. Herkenhoff, 2007, Acquisition using simultaneous sources: 69th
Annual International Conference and Exhibition, EAGE, Extended Abstracts, B006.
Strohmaier, E., J. Dongarra, H. Simon, and M. Mueur, 2017, TOP500 supercomputer sites,
http://www.top500.org/, accessed March 10, 2017.
Verliac, M., V. Lesnikov, and C. Euriat, 2015, The Rousse-1 DAS VSP experiment — Observations and
comparisons from various optical acquisition systems: 85th Annual International Meeting, SEG,
Expanded Abstracts, 5534–5537, https://doi.org/10.1190/segam2015-5886544.1.
Wu, H., W.-F. Wong, Z. Yang, P. B. Wills, J. L. Lopez, Y. Li, B. Blonk, B. Hewett, and A. Mateeva,
2015, Dual-well 3D vertical seismic profile enabled by distributed acoustic sensing in deepwater
Gulf of Mexico: Interpretation, 3, SW11–SW25, https://doi.org/10.1190/INT-2014-0248.1.
Zhan, G., J. Kommedal, and J. Nahm, 2015, VSP field trials of distributed acoustic sensing in Trinidad
and Gulf of Mexico: 85th Annual International Meeting, SEG, Expanded Abstracts, 5539–5543,
© 2017 SEG Page 5260

Seismic Processing and Analysis in Service of Interpretation
Gregory A. Partyka* , OpenGeoSolutions Inc.
Introduction conducive to cookie-cutter, exactly repeatable workflows.

It requires knowledge, experience, and creative, non-linear
Seismic remote sensing reveals structural, stratigraphic and problem solving.
rock mass context. With careful processing and analysis, it
can expose morphology and heterogeneity at scales from Signal Analysis and Shaping / Noise Analysis and
regional to local and can serve as a near-ideal backdrop for Attenuation
subsurface integration and modeling. It is not totally ideal
because seismic resolution continues to be a first order Every dataset exhibits distinct and non-stationary
challenge. Seismic reflections are bandlimited/tuned discrepancies with respect to signal and noise. Data quality
representations of the subsurface. Seismic, in isolation, issues that could detract from the interpretability of the data
cannot expose details that require higher-than-available must be identified. Frequency and wavenumber are
frequency content, or resolve very large-scale architecture examples of dimensions that can help expose and isolate
that requires lower-than-available frequency content. Noise noise trains and footprints so they can be understood and
and weak impedance contrasts further complicate the issue. attenuated. If the noise cannot be isolated, attempts to
attenuate it can lead to degradation of signal. Trade-offs
For seismic information to be used effectively, signal and
must be considered.
noise content must be analyzed, prepared, understood, and
presented in a way that facilitates problem-solving within a
Today, we carry both raw and a denoised versions of the
multi-disciplinary context. Seismic processing and analysis
data through interpretive processing, so that we can
in service of interpretation requires:
compare the effects on attribute. In the future, near real-
(1) signal analysis and shaping / noise analysis and
time processing speed will allow a wider range of denoise
attenuation.
applications to be tested all the way through the
(2) decomposing the data into relevant attributes that
characterization, investigation and interpretation stages.
reveal content.
(3) grouping of attribute types into sets that help
The amplitude spectrum of the input data is a product of the
answer specific questions.
acquisition system and processing scheme. During data
(4) painless data access, presentation, and
preparation, it is critical to shape the spectrum to make
visualization.
maximum use of the signal bandwidth. The desired spectral
(5) investigation and interpretation.
shape depends on what we are trying to do with the seismic
information. As part of investigation and interpretation, we
Today, we routinely perform all of the above, but typically
need to examine both reflection boundaries and layering.
do so after the data has been imaged and reduced to a
To examine reflection boundaries, we need to prepare a
manageable size. As compute and algorithmic capabilities
version of the data with a blue (richer in higher
continue to expand (e.g. speed, memory, machine learning,
frequencies) spectral shape. To examine layering, we need
deep neural networks, etc....), processing and imaging cycle
to prepare data with a red (richer in lower frequencies)
time will reduce to near real-time. This will allow a wider
spectrum. To facilitate imaging geologic content across the
range of parameterization at all stages of seismic analysis
usable bandwidth on a frequency by frequency basis, we
and imaging - from field gathers through post-stack, and
need to remove the spectral colour and prepare a version of
will let us gauge effects of those choices on the
the data with a white spectrum.
interpretation. This will significantly improve image
quality and will help bracket uncertainty.
Source wavelet phase needs to be established to enable
characterization with respect to harder vs softer material.
It is the investigation and interpretation step that is
Wavelet characteristics often vary from well-tie to well-tie
particularly problematic as far as automation. Investigation
and from zone to zone. Acquisition and processing artifacts
and interpretation is not something that is done in isolation
in both seismic and well data (e.g. washouts, tool problems,
with the seismic data alone. It needs to be worked in
etc...) degrade the quality of derived wavelet properties.
context of local area petrophysics, an evolving stratigraphic
Extracting information at multiple wells and multiple
and structural framework, and an evolving static and
zones/windows provides insight regarding uncertainty and
dynamic model of the target zone. The multi-disciplinary
stability. Wavelet estimation is often manually intensive
database of information is never complete and the data
and time consuming. It will remain so in the future.
itself is always less than perfect. Such integration is not
© 2017 SEG Page 5261

Decomposing the Data into Relevant Attributes That time and effort would be much better spent examining
and investigating the geological content. Also, within the
This is now done mainly post migration, but as cycle time team, there is often a wide range of interpretation
continues to improve, it will be done during all stages of experience, knowledge and skills. Not everyone will
processing and imaging to investigate effects of processing always be equally proficient with seismic visualization and
flow choices and parameterization and to ensure optimum the available tools. For reconnaissance scoping to be
imaging of target zones. successful, the information must be presented in a way that
facilitates interaction. Access and visualization must not
For an attribute to be relevant, it needs to be part of a overwhelm those who are not experts in seismic
complementary set that characterizes meaningful rock mass interpretation. At the reconnaissance stage, detection of
properties and facilitates recognition and characterization anomalies is far more important than resolution of details.
of distinct content. It is important to consider which Incorporating an element of 'motion' into the workflow
attributes preferentially respond to which types of becomes extremely helpful. A subtle anomaly which may
anomalies (signal, structure, stratigraphy, rock-mass, fluids, be invisible in a static image often becomes obvious, when
layers, boundaries, side-lobes, etc...). For example, spectral motion/animation is introduced. Techniques such as RGB
decomposition is a good choice for unraveling tuning blending, semi-transparent overlays, principal component
effects. It moves detection and resolution out from under analysis, and classification, can help to consolidate the
the control of the source wavelet and allows impedance and available information and to enhance delineation of
thickness interference to be examined with respect to signal structural and stratigraphic features. The goal is to ensure
and noise on a frequency-by-frequency basis. Spectral nothing of importance is missed, and to efficiently high-
inversion is a better choice when it comes to defining grade the content for follow-up detailed studies.
layering architecture. It maximizes the resolution and
detection of softer versus harder material via the available We are currently in a transition time. Technology and speed
signal bandwidth. For defining structure, multi-scale associated with data access, presentation and visualization
curvature is a good first choice. It provides an objective and is presently not fast enough to fully automate this aspect of
data-driven view of the structural framework. Discontinuity the workflow. In the future, automation will bypass such
attributes are most helpful for imaging and exposing edges functional issues and will allow you to step directly into
and texture. Elastic inversion facilitates rock versus fluid examining and investigating the geological content.
identification. It enables direct computation of band limited
elastic parameters such as density and shear modulus. Grouping of Attribute Types
Some attributes can be generated in a highly automated Investigation and interpretation require keeping track of
manner. Other attributes such as elastic inversion require many dimensions (e.g. X, Y, time, depth, frequency, spatial
significant attention to parameterization and require scale, offset/angle, wavelet shape, boundaries, layers,
substantial user intervention. This makes them highly architecture, etc....). We are pretty good at keeping track of
dependent on not only the quality of the seismic data, but one, two, or three dimensions, but adding a fourth or fifth
also the quality of the well information, and the human dimension starts to stress human capabilities and also the
guidance required to integrate the usable content. capabilities of today's interpretation technology. Current
interpretation workstation technology makes it reasonable
Painless Data Access, Presentation, and Visualization for those who are skilled and experienced in the art of
seismic investigation and interpretation to sort through the
Painless data access, presentation, and visualization is just dimensionality and uncertainty in a reasonably systematic
as important as the content itself. The easier it is to probe way. Artificial intelligence will help us navigate through
the data, the more likely important features will be the dimensionality in a more rigorous way. To cope with
recognized and localized. In today's world, the database can the many dimensions, we often break up the processing and
easily become overwhelming. There are many aspects to characterization tasks into smaller manageable pieces.
this problem. These include the sheer volume of computed
attributes, problems with data access, and problems with Three particularly useful data forms that are often grouped,
data visualization. If the data and access to the data are not are: blueing (bandlimited reflectivity), coloured inversion
managed reasonably, the interpreter is forced to focus on (bandlimited impedance), and spectral inversion
functional issues such as: (subsurface architecture confidence). As a set, these three
• Which attributes are going to be useful? provide complementary insight that is substantially more
• Where in x, y, t is the information? informative than a single view of the data. Together, they
enable the characterization of the subsurface content within
• Which volumes do I load?
the usable signal bandwidth. While blueing and coloured
• How do I load the volumes? inversion reveal the same geological content - one as
• How should I scale the data? impedance boundaries and the other as impedance layers,
• What colour bar should I use? spectral inversion reveals the underlying subsurface
architecture. In the case of blueing and coloured inversion,
© 2017 SEG Page 5262

the usable signal bandwidth can be shaped to maximize lead to multiple working hypotheses. As always, there will
resolution at the expense of side-lobes or minimize side- unfortunately continue to be a desire to find 'silver bullet'
lobes at the expense of some resolution. Side-lobe effects automated solutions (e.g. direct hydrocarbon indicator,
need to be modeled and considered when interpreting direct fracture density indicator, etc....), but the whole idea
subtle details. This becomes particularly important for of a 'silver-bullet' runs counter to what really needs to take
target zones with small impedance contrasts that are place. Activities related to investigation and interpretation
adjacent to large impedance contrasts. Spectral inversion must remain open in scope. They should lead to exciting
transforms spectral interference to layering patterns. It "a-ha" moments and lead to comments such as "well that's
characterizes subsurface architecture in a way that differs interesting" or "I didn't expect that". Hearing these phrases,
from blueing and coloured inversion. Blueing and coloured signals learning, a shift in preconceived notions, and
inversion link resolution and detection to the source improved understanding of the solution space. Future
wavelet shape and to attributes such as tuning thickness and technology breakthroughs will add a significant amount of
tuning frequency. In the case of spectral inversion, the interpretation 'safety' feedback that will help interpreters to
source wavelet overprint is removed as part of the stay within plausible solution spaces, to more quickly
processing. This leads to a concept of resolution and understand the unexpected and anomalous behavior, and to
detection that is no longer tied to source wavelet shape help constrain and iterate away from unreasonable, towards
attributes. The ability to detect and/or resolve becomes a more probable solutions. Cycle time will continue to
function of the available signal bandwidth, signal-to-noise decrease, but there will always be a need for investigation
ratio, the complexity of the local layering architecture, and and interpretation think-time in context of experience and
the complexity of impedance contrasts within the local skill.
distribution of layering. A simpler layering architecture and
impedance distribution requires less bandwidth to achieve Just as it is helpful to group attribute types into sets that can
resolution and detection. During interpretation, spectral help answer specific questions, it is likewise helpful to
inversion, blueing and coloured inversion make for a group interpretation goals into manageable pieces that are
powerful combination that enables characterization and not overwhelming. One such approach is to split
maintains integrity in context of complementary evidence. investigations into (1) reconnaissance scoping that ensures
The tuned response in blueing and coloured inversion early awareness of the geological content and anomalies in
should not contradict the details exposed by spectral the seismic expressions, and (2) targeted detailed
inversion. analysis/interrogation. Once anomalies and interesting
features are identified, workflows can be outlined to further
Just as it is helpful to pair the complementary information interrogate the data for more comprehensive understanding
that comes from blueing and coloured inversion (one of the details. Combining forward modeling (i.e. modeling
reveals boundaries - the other reveals layers), it is likewise seismic expressions of plausible geological scenarios) and
helpful to pair discontinuity attributes (that inverse problem-solving (i.e. analysis and geological
reveal boundaries and texture) with attributes that describe characterization of observed seismic expressions) leads to a
what happens within target bodies. For example, better understanding of uncertainty and evaluation of
discontinuity serves as a valuable backdrop and context for multiple working hypotheses. Modeled seismic properties
the tuning information that is revealed by spectral rarely match all aspects of real seismic data. Similarities
decomposition. Discontinuity likewise pairs nicely as a strengthen calibration. Discrepancies reveal inadequacies in
backdrop for geometric flexuring information that is the data, conceptual framework, characterization, and/or
revealed by multi-scale curvature. In both cases, pairing of mathematical models. Learnings from such detailed
multiple attributes provides insight that is greater than the analyses can then be integrated back into the database for a
sum of the individual parts. new round of reconnaissance scoping.
Investigation and interpretation The earlier in the processing stream that seismic,
petrophysical, geological and reservoir data can be
It is the investigation and interpretation step that is compared and calibrated, the more likely a reasonable
particularly difficult to automate. This is because reservoir flow-unit and storage-unit characterization will
investigation and interpretation is not done with seismic evolve. If integration is delayed and attempted after
data in isolation. It needs to be done in context of local area completing the petrophysical characterization, and after
petrophysics, a stratigraphic and structural framework, and seismic data is processed and handed off to the interpreter,
an evolving static and dynamic model of the target zone. the integration becomes more rigid; dependent on the
assumptions built into both the petrophysics and seismic
Such integration requires and will continue to require think sides of the investigations. Problems with seismic data
time (deeper thinking) and will continue to rely on the processing and petrophysical data processing become most
geophysicist's experience and knowledge of not only the obvious during the integration stage, when mismatches
seismic piece of the puzzle, but also the larger multi- between petrophysical information and geophysical
disciplinary controls. It will likewise continue to require information become evident. If such integration (and
consideration of non-unique and uncertain indicators that problem detection) is done at a late project stage, the
© 2017 SEG Page 5263

interpreter can either force the data together or backup and The real silver-bullet is the mind of the interpreter who is
reprocess the data. In each case, time and quality problem- open to possibilities and who is armed with non-stifling
solving is lost. tools.
In the future, the interpreter will be able to fix seismic Acknowledgments

processing issues in real-time, as they are found. Future

technology development will leverage the full The author thanks the individuals at OpenGeoSolutions
dimensionality of the data in a way that will gently yet who as a team are forging down this road-ahead and the
objectively guide investigations and integration in a parallel customer community for their constant support and
manner. This will help flag anomalies and will help feedback.
question hypotheses. It will allow us to interact with the
multi-dimensional solution space, identify probable
outcomes, end-members and degrees of uncertainty. The
future interpretation system will make use of the full
dimensionality in real time to generate an architecture,
within which any identified anomaly can be considered
(investigated and interpreted) in context of all processing
choices and the larger context of petrophysical, geological,
and reservoir characteristics.
Summing Up
There is a wealth of geological insight in seismic data. To

maximize the information that can be extracted requires
careful preparation and analysis. It also requires real-time
presentation and visualization that avoids problems
associated with data access and data management.
Hardware and software development will enable such real
time investigations. It will not however, change the fact
that in the world of remote seismic sensing, processing and
analysis in service of interpretation, is and will continue to
be, interpretive with many inherent uncertainties that need
to be considered in the larger petrophysical, geological and
reservoir context.
There are also many technological and human issues to

consider. The “great crew change” has largely been
accomplished through the commodity downturn of 2015-
16. Within the group that is still employed, there are now
serious gaps in interpretation experience, knowledge and
skills. There are fewer experienced mentors, and more is
expected from the remaining knowledgeable practitioners
who will become more and more reliant on machine help.
The tech-savvy generation will expect to be guided. In the

future, interpretation systems will have more built-in
expertise, common sense defaults, and artificial intelligence
that will learn across an organization or perhaps even
across organizations. Silver-bullets will continue to be
sought after. Increasingly powerful tools will be provided
to less experienced hands. This could lead to a loss of
individual expertise (very dangerous) which will in turn
enact pressure to simplify analyses to a common (i.e.
lower) expertise denominator. Despite all the new
technology there will always be oil and gas to be found
outside of the automation.
© 2017 SEG Page 5264

EDITED REFERENCES
REFERENCES
Al-Dossary, S., and K. J., Marfurt, 2006, 3D volumetric multispectral estimates of reflector curvature and
rotation: Geophysics, 71, no. 5, P41–P51, https://doi.org/10.1190/1.2242449.
Kallweit, R. S., and L. C. Wood, 1982, The limits of resolution of zero-phase wavelets: Geophysics, 47,
1035–1046, https://doi.org/10.1190/1.1441367.
Partyka, G., 1998, Seismic petrophysics integration,
http://www.freeusp.org/RaceCarWebsite/TechTransfer/OnlineTraining/SeisPetro/SeisPetro.html.
Partyka, G., 2015, Deriving subsurface insight from seismic: Abstract from the Indonesian Petroleum
Association Thirty Ninth Annual Convention & Exhibition.
Partyka, G., 2005, SEG Distinguished Lecture, http://www.seg.org/education/lectures-
courses/didstinguished-lecturers/spring2005/partykaabstract.
Partyka, G., 2016, Subsurface insight from seismic attributes: Abstract from the 2016 Geophysical
Society of Houston Spring Symposium.
Steindl, J., Nilssen, I. R., Askim, O. J., O. Barkved, G. Partyka, 2015, Use of quantitative seismic analysis
to define reservoir architecture and volumes — An example from Johan Sverdrup Field: Abstract
from the Norwegian petroleum society conference on reservoir characterization.
Widess, M. B., 1973, How thin is a thin bed?: Geophysics, 38, 1176–1180,
https://doi.org/10.1190/1.1440403.
© 2017 SEG Page 5265

EDITED REFERENCES
REFERENCES
Al-Dossary, S., and K. J., Marfurt, 2006, 3D volumetric multispectral estimates of reflector curvature and
rotation: Geophysics, 71, no. 5, P41–P51, https://doi.org/10.1190/1.2242449.
Kallweit, R. S., and L. C. Wood, 1982, The limits of resolution of zero-phase wavelets: Geophysics, 47,
1035–1046, https://doi.org/10.1190/1.1441367.
Partyka, G., 1998, Seismic petrophysics integration,
http://www.freeusp.org/RaceCarWebsite/TechTransfer/OnlineTraining/SeisPetro/SeisPetro.html.
Partyka, G., 2015, Deriving subsurface insight from seismic: Abstract from the Indonesian Petroleum
Association Thirty Ninth Annual Convention & Exhibition.
Partyka, G., 2005, SEG Distinguished Lecture, http://www.seg.org/education/lectures-
courses/didstinguished-lecturers/spring2005/partykaabstract.
Partyka, G., 2016, Subsurface insight from seismic attributes: Abstract from the 2016 Geophysical
Society of Houston Spring Symposium.
Steindl, J., Nilssen, I. R., Askim, O. J., O. Barkved, G. Partyka, 2015, Use of quantitative seismic analysis
to define reservoir architecture and volumes — An example from Johan Sverdrup Field: Abstract
from the Norwegian petroleum society conference on reservoir characterization.
Widess, M. B., 1973, How thin is a thin bed?: Geophysics, 38, 1176–1180,
https://doi.org/10.1190/1.1440403.
© 2017 SEG Page 5266

Inversion of seismic data for reservoir properties: recent advances and the road ahead
Patrick Connolly*, PCA Ltd
Summary uncertainties from mapping impedances to reservoir

properties, while ignoring others such as the uncertainty of
The industry is transitioning from deterministic to the low frequency model.
probabilistic seismic inversion methods. The change is
occurring slowly and, so far, has been only partial. The At the highest level we can envisage the characteristics of
requirement is clear but the challenges of developing an ideal inversion process. It should combine as much
practical, fully probabilistic seismic reservoir prior information as possible with the seismic; the
characterization workflows are significant. uncertainty of all data should be assessed and accounted for
in the process and the output should in some sense be
probabilistic.
Introduction
Ultimately all uncertainty estimates are subjective; there is
Geophysicists have been making estimates of reservoir no way to measure all possible sources of uncertainty; there
properties from seismic data for decades and, with ever is no ‘right’ answer. Our objective must be to develop
improving data quality, a deeper theoretical understanding applications that allow a geoscientist to combine all the
and better algorithms, there has been a steady increase in relevant information about a reservoir. This may take the
reported successes. Mostly unreported, of course, are the form of ‘hard’ well data to calibrate, for example, a rock
failures. physics relationship or softer probabilities such as
expectations of facies proportions based on a geological
The interpreting geologist or geophysicist faces a difficult model. It will be their judgement as to what is important
problem in determining how much confidence to have in and what the uncertainties are.
predictions based on seismic inversion. An experienced
processor or interpreter can judge the quality of a Workflows need to be practical. Run-times must be
conventional seismic section but can find it difficult to manageable and applications must be accessible. The user
determine the reliability of inversion results (other than by should be focused on the geology rather than on fine-tuning
drilling). To improve decision making, the industry a list of meaningless parameters. And, above all, processes
requires probabilistic estimates which aim to account for all must be transparent; it must be clear to the user what the
the key uncertainties. algorithms are doing. To quote Angela Merkel (slightly out
of context!); ‘Algorithms, when they are not transparent,
Seismic reservoir characterization requires the integration can lead to a distortion of our perception’.
of seismic data with other data types and other sources of
information; well log data, rock physics models, geological Achieving all these aims will not be easy but we are edging
models, spatial statistics and so on. When combining data closer. In this article I’ll briefly discuss progress in some
it is important to take into account their relative of the workflow components and suggest how they might
uncertainty; data with low uncertainty should be weighted be combined to meet our objectives.
up; data with high uncertainty should be weighted down.
Without this we risk biasing the result. If we assign (Note on terminology: I use the description ‘probabilistic’
uncertainties to all the data we can achieve the best balance rather than ‘stochastic’, which is sometimes used to
between the prior and seismic data as well as obtaining a distinguish algorithms from deterministic methods. A
probabilistic result. This is the essence of Bayes theorem. stochastic optimization algorithm, such as simulated
annealing, may result in a single, essentially repeatable,
Probabilistic seismic inversion algorithms based on Bayes output; the usual definition of deterministic. Describing an
theorem emerged in the early 2000’s (Buland and Omre, algorithm as ‘stochastic’ is not necessarily saying very
2003; Gunning and Glinsky, 2004). Progress has continued much.)
mostly within academia and oil companies (Leguijt, 2009;
Grana and Della Rose, 2010; Riise et al, 2012; Connolly
and Hughes, 2016) but we have yet to see broad uptake Components of a probabilistic inversion workflow
across the industry.
Most inversion algorithms are based on ‘classical’ AVO;
Probabilistic elements have been introduced into existing methods that are post-imaging, with velocity and amplitude
commercial packages but they tend to be partial resulting analysis as separate steps. Classical AVO is ray
in, say, facies probability volumes that account for the theoretical, based on the Zoeppritz equations or their
© 2017 SEG Page 5267

Inversion of seismic data
linearizations and utilizes plane wave reflection coefficients error is the scaling of the gradient term, whether explicit or
and a convolutional model. Anisotropic modifications can implicit, caused by incorrect pre-stack scaling, ignored
be introduced but, other than for azimuthal analysis (not anisotropy, incorrect velocities, move-out errors and so on.
considered here), typically they are not. If the gradient is being combined with the intercept as a
chi-projection the gradient error will show up as an error in
This field has reached a high level of maturity in recent the actual chi angle. Therefore, if coordinate rotated
years. Theoretical understanding is strong, different seismic is being utilized as input to the inversion a large
approaches have been reconciled and we have a clear component of the seismic error could be captured as an
understanding of possibilities and limitations. For uncertainty in the chi angle.
example, I believe there’s a broad consensus that in
general, making useful estimates of the third AVO term, Rock physics provides the link between elastic and
the curvature, is unrealistic. Beyond about 30⁰, amplitudes reservoir properties. Although fully theoretical models are
can change rapidly with angle such that, at the very least, not available this nonetheless is a mature discipline and the
implausibly accurate, high resolution interval velocity current pragmatic mix of theoretical, heuristic and
models would be required to define the offset to angle empirical models appears to work well (Avseth et al, 2005).
relationship. The sub-discipline of statistical rock physics provide ways
to capture the uncertainty of rock property relationships.
Within these limitations much is possible; combinations of
intercept and gradient correlate with a number of elastic There is increasing recognition that direct, one-step
properties to a high degree of accuracy as demonstrated by inversion to reservoir properties is preferable to a two-step
the  approach (Goodway et al, 1997) and, semi- process of estimating impedance and then mapping
empirically, via extended elastic impedance (Whitcombe et impedance to reservoir properties. The former allows the
al, 2002). More recently, the development of relative rock integration of a wider range of geological data and
physics (Ball et al, 2014) has provided a framework to knowledge (Bosch et al, 2010; Kemper and Gunning,
express relative impedance properties and relative elastic 2014).
properties in the same form allowing their relationships to
be made explicit. Most inversion algorithms still rely on the use of low
frequency models (LFMs) which are either directly added
The starting point for almost all inversion methods are to the seismic bandwidth component or used as a starting
flattened gathers or directly derived products such as angle point for an optimization algorithm. Sometimes the
stacks or intercept and gradient measurements. If the data uncertainty of the LFM can be low and their use is
are limited to about 30 degrees then, in principle, the appropriate but if, for example, the reservoir thickness is
information content can be captured by the combination of greater than about 50-60ms then the LFM will likely
an intercept and gradient. I think it best to always calculate contain a component related to the reservoir. In this case, if
the gradient explicitly (using a 3-term fit to reduce the uncertainty of the LFM is not accounted for, the
curvature bias) to reveal any inherent problems with the resultant reservoir property estimates will be biased (Grant,
gathers rather than leaving the term implicit. 2013).
There is often a divergence in the workflow at this point. Geology tends to exhibit a degree of lateral continuity so it
One branch is for qualitative AVO based on coordinate would appear sensible to impose lateral constraints on
rotations of intercept and gradient, often followed by results. Geostatistical inversions incorporate a model for
colored inversion, producing datasets that correlate with lateral correlation which allows the simulation of higher
elastic or reservoir properties (Yenwongfai et al, 2017). frequency data; potentially important for building reservoir
For the other branch, gathers or angle stacks are input to an models for subsequent use in dynamic simulation. However
inversion algorithm to estimates absolute impedances there are a number of potential difficulties with this
followed by a mapping to reservoir properties. It may be approach.
better to combine these two options to allow the optimized,
coordinate rotated seismic as input into the next step to Two point geostatistical models often fail to effectively
estimate reservoir properties (Hafez and Castagna, 2016). model real geology and parametrization can be very
uncertain. Multi-point geostatistics has been integrated
As with all components of the inversion, the uncertainty of with inversion but this adds a lot of complexity. There is
the seismic should be estimated and accounted for. This is also a requirement to regrid from Cartesian to stratigraphic
extremely challenging. A simple signal-to-noise value is coordinates; not something to be done too early in the
inadequate (and impossible to measure). If the AVO workflow.
information is being exploited then probably the largest
© 2017 SEG Page 5268

How are uncertainties to be captured? For a 1D inversion, inversion and reservoir modelling is required to determine
with frequencies limited to the top end of the seismic, whether this is a viable option.
uncertainties can be represented as sample-by-sample
facies probabilities or as reservoir property means and
variances. However a high frequency geostatistical Workflow
inversion requires multiple realizations to represent the
uncertainties. Apart from the higher run-times of the Figure 1 shows the outline of a workflow that captures
inversion, much complexity is added to the workflow to most of the points discussed above. It includes some
carry these realizations through subsequent steps. And established methods, some emerging and some yet to be
further iteration may be required to adjust the geostatistical fully realized but I think is close to being achievable with
parametrization to history match a simulation model. existing technology.
Combining the inversion with the geostatistical modelling
places a big overhead inside the loop. The target is for all key uncertainties to be captured and to
provide a probabilistic solution. And it must be practical;
More fundamentally the results of geostatistical inversion is none of the components should have excessive run-times
a mix of real data, the seismic or what amounts to a and the workflow shouldn’t be too complex.
remapping of the seismic, with simulated data. This can
result in a confusing picture; below the maximum seismic The workflow is split into as many separate components as
frequency there is a real image and above this frequency possible allowing each phase of the operation to be
there is a statistical simulation. What is actually being optimized rather than trying to get everything right inside
resolved? An inability to distinguish what is real from one large application. Breaking down the workflow allows
what is simulated may be ideal for a Hollywood sci-fi the user to develop an intuitive feel for the relationship
blockbuster but not necessarily so good for an image of a between input and output. Ultimately all steps need to be
reservoir. tuned by eye; it is not possible to calculate the best
parameters as there are always some unknown variables
A better approach may be to separate the inversion from the that lie outside any model.
geostatistical modelling. First invert in 1D, restricting
frequencies to the high end of the seismic data, and obtain a The workflow begins with the gathers from which the
probabilistic result. Then use this to constrain subsequent intercept and gradient are measured. The quality of the
geostatistical modelling potentially iterating with dynamic gradient is carefully checked and then colored inversion is
simulation. More work on the linkage between seismic applied to obtain relative AI and GI (gradient impedance).
Figure 1: A possible probabilistic seismic reservoir characterization workflow. It is broken down into many seperate components allowing each
to be optimized before progressing to the next. Uncertainties are captured in the output from the one-dimensional inversion. Multiple
realizations are only generated towards the end of the workflow where they can be iterated to history-match with production data.
© 2017 SEG Page 5269

These are combined to obtain relative EEI datasets that did could would they work together harmoniously?) This, I
optimally correlate with desired elastic or reservoir believe, goes someway to explain the slow pace of
properties. The theoretical optimal rotation angle, chi, will progress.
usually be different from the actual seismic chi value
because of the gradient error and so this step needs to be What to do? A starting point might be to try to improve
optimized by eye. communication between disciplines. Typically the authors
of papers in this area approach the subject from a specific
The relative EEI datasets have direct value themselves but point-of-view; as geostatisticians, rock physicists, AVO
can also be used as input into a one dimensional specialists, computer scientists, and so on, (the astute
probabilistic inversion algorithm that integrates them with reader may have detected my background in AVO). This
rock physics models and data, probabilistic expectations inevitably leads to a bias in thinking and a potential failure
from the geological model, vertical statistic and so on. This to fully exploit the potential of each sub-discipline (or,
will produce facies probability volumes and reservoir more often, to appreciate their limitations). Rather than
property estimates. writing as if addressing fellow specialists it may be better
to try to address the broader seismic reservoir
The next step is to regrid these data onto a stratigraphic characterization community. Without turning papers into
coordinate system and use this to constrain geostatistical tutorials perhaps we could be a little more generous with
modelling. If this modelling step is restricted to having explanations to allow our writing to be more accessible.
minimal impact within the seismic frequency band it will
be consistent with the seismic and the previously included
prior data. The modelling can then be iterated within a Conclusions
dynamic simulation loop to history match with production
data. The industry is transitioning from deterministic to
probabilistic seismic inversion methods. The challenges of
This represents in some respects a radical departure from a developing practical workflows are formidable and
the existing standard approach; combining seismic with progress has been a little slow. The recent development of
low frequencies, inverting to absolute impedance and more complete Bayesian reservoir characterization methods
mapping to reservoir properties. is encouraging. An ability to routinely obtain probabilistic
property estimates would be significantly beneficial to the
This workflow still has some significant limitations, not industry.
least those inherent in classical AVO. Further significant
improvements will require the integration of velocity and
amplitude analysis perhaps in some form of hybrid scheme
with elastic FWI or perhaps with FWI directly providing
the characterization results (Naeini et al, 2016).
The road ahead
What is holding up the development of new workflows and

particularly of probabilistic methods? Let’s return to the
list of data that we wish to integrate and in particular
consider the implied range of expertise required by
developers and, to some extent, users of seismic inversion.
As our aim is to estimate geological properties the first

requirement for practitioners is some level of geological
knowledge. Seismic data processing and petrophysics
experience is needed to ensure an understanding of the
input data. Some expertise is essential in AVO, rock
physics, geostatistics, Bayesian inference, computer science
and probably yet more. Few individuals can claim such a
breadth of knowledge and even large oil companies, service
companies and academic institutions struggle to build
multidisciplinary teams to cover all these skills (and if they
© 2017 SEG Page 5270

EDITED REFERENCES
REFERENCES
Avseth, P., T. Mukerji, and G. Mavko, 2005, Quantitative seismic interpretation: Cambridge University
Press, http://dx.doi.org/10.1017/CBO9780511600074.
Ball, V., L. Tenorio, J. P. Blangy, C. Schiott, and A. Chaveste, 2014, Relative rock physics: The Leading
Edge, 33, 276–286, http://dx.doi.org/10.1190/tle33030276.1.
Bosch, M., T. Mukerji, and E. F. Gonzalez, 2010, Seismic inversion for reservoir properties combining
statistical rock physics and geostatistics: A review: Geophysics, 75, no. 5, 75A165–75A176,
http://dx.doi.org/10.1190/1.3478209.
Buland, A., and H. Omre, 2003, Bayesian linearized AVO inversion: Geophysics, 68, 185–198,
http://dx.doi.org/10.1190/1.1543206.
Connolly, P. A., and M. J. Hughes, 2016, Stochastic inversion by matching to large numbers of pseudo-
wells: Geophysics, 81, no. 2, pp. M7–M22, http://dx.doi.org/10.1190/geo2015-0348.1.
Goodway, B., C. Taiwen, and J. Downton, 1997, Improved AVO fluid detection and lithology
discrimination using Lamé petrophysical parameters “λρ”, “μρ”, & “λ/μ fluid stack”, from P and
S inversions: 67th Annual International Meeting, SEG, Expanded Abstracts, 183–186,
http://dx.doi.org/10.1190/1.1885795.
Grana, D., and E. Della Rossa, 2010, Probabilistic petrophysical-properties estimation integrating
statistical rock physics with seismic inversion: Geophysics, 75, no. 3, O21–O37,
http://dx.doi.org/10.1190/1.3386676.
Grant, S. R., 2013, The impact of low frequency models on reservoir property predictions: 75th Annual
http://dx.doi.org/10.3997/2214-4609.20130335.
Gunning, J., and M. Glinsky, 2004, Delivery: An open-source model-based Bayesian seismic inversion
program: Computers and Geosciences, 30, 619–636,
http://dx.doi.org/10.1016/j.cageo.2003.10.013.
Hafez, A., and J. P. Castagna, 2016, Distinguishing gas-bearing sandstone reservoirs within mixed
siliciclastic-carbonate sequences using extended elastic impedance: Nile Delta — Egypt:
Interpretation, 4, no. 4, pp. T427–T441, http://dx.doi.org/10.1190/INT-2015-0223.1.
Kemper, M., and J. Gunning, 2014, Joint Impedance and Facies Inversion– Seismic inversion redefined:
First Break, 32, no. 9, 89–95.
Leguijt, J., 2009, Seismically constrained probabilistic reservoir modelling: The Leading Edge, 28, no.
12, 1478–1484, http://dx.doi.org/10.1190/1.3272703.
Naeini, E. Z., T. Alkhalifah, I. Tsvankin, N. Kamath, and J. Cheng, 2016, Main components of full-
waveform inversion for reservoir characterization: First Break, 34, no. 11, 37–48,
http://dx.doi.org/10.3997/1365-2397.2016015.
Riise, O., J. Elgenes, J. M. Frey-Martinez, Ø. Kjøsnes, and A. Buland, 2012, Detailed lithology and fluid
mapping of the asterix gas discovery using bayesian inversion methodology: 74th Annual
http://dx.doi.org/10.3997/2214-4609.20148160.
Whitcombe, D. N., P. A. Connolly, R. L. Reagan, and T. C. Redshaw, 2002, Extended elastic impedance
for fluid and lithology prediction: Geophysics, 67, 63–67, http://dx.doi.org/10.1190/1.1451337.
© 2017 SEG Page 5271

Yenwongfai, H. D., N. H. Mondol, J. I. Faleide, and I. Lecomte, 2017, Prestack simultaneous inversion to
predict lithology and pore fluid in the Realgrunnen Subgroup of the Goliat Field, southwestern
Barents Sea: Interpretation, 5, no. 2, SE75–SE96, http://dx.doi.org/10.1190/INT-2016-0109.1.
© 2017 SEG Page 5272

High-performance computing for seismic imaging; from shoestrings to the cloud
Sverre Brandsberg-Dahl, PGS
with the ever-increasing computational power, see Figure

Introduction 1. The investment levels at the time in such hardware must
have been substantial, and some of the biggest commercial
The process of creating an image of the Earth’s interior HPC installations from then until present day have been
from surface seismic measurements has always required dedicated to seismic imaging. Around the mid 2000’s
some sort of computation. Norm Bleistein was the first to compute clusters based on commodity hardware were
impressed on me the richness in approaches and methods to introduced to the field, and rapidly concurred all aspects of
solving this rather complex problem, back in the mid 90’s seismic imaging. By connecting together hundreds or even
when I was a young graduate student. Ten years earlier, thousands of individual computers, pretty much anyone
Bleistein (1987) had introduced to the field of geophysics, a could build and have access to supercomputer power.
rigorous treatment of Kirchhoff migration through the tools Clusters have remained the mainstay of the industry until
and language of asymptotic analysis. This provided a today. The latest addition to this trend is the move to more
roadmap for how one could implement in a computer, an use of cloud computing, where cluster computing is offered
imaging algorithm that would yield true-amplitude images as a service by a variety of providers.
of reflectors in the Earth. However, at this point in time,
Norm was also truly fascinated by how geophysicists had According to the trend shown in Figure 1, the seismic
discovered many of these principles years earlier, mainly industry should soon be entering the era of exascale
through ingenuity and a deep understanding of the physics computing. When this will happen is still highly uncertain,
behind the seismic experiment and how to map the given market forces and reduced investment levels, but also
reflective layers in the sub-surface using reflections. due to technical challenges facing the scalability of seismic
The earliest imaging algorithms had solved the problem of imaging algorithms. Until today, it has been pretty much a
mapping reflectors in the subsurface, not through hunt for Flops, but dataset sizes and input/output are maybe
asymptotic analysis, but by using a string and pencil. One now the biggest obstacles to continuing on the trend from
of the earliest examples of this can be found in the mapping the past. Also, more complex, heterogeneous computer
“computer” of Haggerdorn (1958). Together with a fellow architectures, are introducing challenges around
student at the time, we actually constructed such a programmability, so it might not be as easy now to adopt
“computer”, not using a single transistor, but rather new algorithms to the hardware of the day as in the past.
shoestring, nails and a pencil. With this we were able to There is an important lesson to be had from the now
map the classic “seismic bowtie” from a common offset vintage machines of the 90’s, in that they were working
section into its corresponding syncline in the subsurface with a nicely balanced ratio of I/O to compute. I will come
space. back to this later when discussing some of the trends we
might face in the years ahead.
Following such early innovations, the groundbreaking work
of John Claerbout (1971, 1985) caused nothing short of a
seismic shift in the way industry and academia approached
the subject of reflector mapping or migration. With the
introduction the exploding reflector principle and a series
of new and innovative ways of imaging subsurface
reflectors, the computational era has truly arrived for the
field of seismic imaging. Rapid innovation by the industry
and academia introduced several innovative and robust
approaches to how one could use a computer to effectively
create reflector maps from surface seismic data. From these
very early days until the 90’s, this state of the art seismic
imaging was using the available state of the art computers
to perform this task. That meant that seismic imaging was
being performed using mainframes and the many different
supercomputers of the era. Computers from Thinking Figure 1: The evolution of seismic imaging algorithms and
Machine, SGI, IBM and Cray were all commonplace as the relative compute power availabiltiy. The zoom shows some
tools of choice for seismic processing and imaging. The of the most powerful supercomputers of the last 5 years.
seismic companies and oil companies alike were big (taken from a presentation by Calandra et al., 2017, Rice
customers of high end computing equipment, and the Oil & Gas HPC Conference)
imaging algorithms being developed were evolving along
© 2017 SEG Page 5273

HPC for seismic imaging
Breaking with a long standing paradigm; clusters acquired as individual shot records, is recorded as
continuous records capturing multiple shots. Such records
Seismic imaging went through a paradigm shift about 20 do no lend themselves to the classical processing schemes
years ago when the first compute clusters were introduced. of cluster-based RTM, unless some form of data de-
This caused a rapid and decisive shift in industry compute blending is performed (Berkhout, 2009). To treat the
platform; away from the earlier generations of mainframes continuous records directly is putting further pressure on
and supercomputers to relatively cheap clusters built using compute architecture and algorithms, in that the time axis
commodity components. now becomes much longer. An illustration of this is shown
On the algorithmic front, this helped facilitate the shift to in the bottom part of Figure 2. This also shows the back
shot based migrations, where rather than treating the data in pane of a Cray XC30 computer, where the massive wire
CMP order or as offset classes, the seismic data was simply bundles represent the network infrastructure provided to
imaged one shot at the time. This ushered in the era of shot support the shared memory architecture of this machine.
migration and introduction of the term “embarrassingly With sufficient memory, one can imagine that the complete
parallel” application, as scalability was achieved by simply seismic experiment can be treated as one, offering hope for
adding more computers to do more shots at a time. An moving beyond the reliance on individual treatment of shot
illustration of data and some legacy hardware is shown in records that prevail in most algorithms today. The choice of
the top portion of Figure 2. Each shot could be handed off algorithms closely follows the trends in available hardware,
to an individual computer, migrated ‘in the box’, and then so while the clusters have reigned supreme for quite some
the image is formed by collecting all the individual sub- time, there might be a change on the horizon towards new
images into a final stack/image. The fidelity of the algorithms and infrastructure to support them.
algorithms, and problem size (aperture and frequency) was
then basically tracking the evolution of improving CPU
speed and increasing memory size. This trend is also
observable on the graph shown in Figure 1. As long as the
individual shot fit inside the footprint of the individual
server, this was indeed embarrassingly parallel, and
enabled a rapid growth in capacity and capability of
imaging algorithms. With the introduction of GPUs this
trend continued, and even more compute flops were
available to solve the imaging problems faster or more
accurately.
It is not until recent days that this model has started to be

challenged, mainly due to the slower growth in memory
size and performance. As larger and more finely sampled
computational domains are required, applications like RTM
started to push at the boundaries of what is physically
possible to do inside a single server. This trend was also
helped along, or maybe pushed along, by the ever more
advanced seismic surveys and larger data volumes acquired
by the industry. The introduction of very long offsets in
marine seismic (Long et al., 2014) pushed the problem size
up by a factor 4x as offsets increased from a typical 8km to
more than 16km and in a full-azimuth fashion, requiring a
doubling of a typical computational aperture, just to capture
the input data. Computational performance for algorithms
like RTM is obviously still key, but the true hotspot when
Figure 2: A comparison of data structure and
imaging such datasets is now shifting to lack of available
address space (memory) to efficiently hold the problem computaational models for the classical shot-parallel data
inside the computer. With this backdrop, a logical and hardware (top), as compared to continous data
challenge to the prevailing shot parallel implementations recordings with blended acquisition, and an example of
was introduced, and pushed us to look for solutions that compute hardware to support the use of shared memeory, a
could provide access to very large memory footprints. Cray XC30 system.
This move is also supported by the rapid adaptation of
blended data acquisition; where data instead of being
© 2017 SEG Page 5274

Implementing RTM on a shared memory architecture Continued analysis of performance of the now distributed
FFT-based implementation showed that further
In this section, I will review the effort undertaken by a team optimization was possible by rearranging some of the
of geophysicists and computer engineers to port a RTM computational kernels of the code as well as optimizing the
code originally developed to run on a cluster to run on a transposes used when applying the TTI anisotropic terms in
shared memory architecture machine. The RTM code that the extrapolator. After these items were addressed, the
was originally designed to run on dual-socket servers with relative run time for the distributed code on XC40 was less
local scratch disk for storage of snapshots, and the goal is than when running one shot per node. Surprising maybe,
to port it to run on a Cray XC40 machine. The XC40 is also but this reflect the fact that snapshots are now stored in
built around dual socket compute nodes, but no local memory with much faster access than even what was
scratch is available, and the nodes are networked to allow provided by high-performance, in node scratch disk.
for rapid access also to memory not local to any node. The
same CPUs were used in both platforms; so one could
maybe easily conclude that the best possible outcome
would be parity between running a shot inside a single node
in the cluster versus across several nodes on the XC40.
The first step in this effort was to simply run the code, as is,
on the XC40. As there is not local scratch disk in the nodes,
this would rely on storing the snapshots on a centralized,
parallel file system instead. The relative performance of
this is shown in Figure 3, indicating that such a straight port
resulting in about a 3x slowdown. The next step was to
mitigate the lack of local disk. This was done by instead
leveraging the very large memory pool available in the
XC40 for storing the snapshots. This obviously required the
use of multiple compute nodes per shot migration to create Figure 3: Relative performance of RTM when run fully
a sufficient memory footprint to hold both the inside a dual socket commodity node and on a shared
computational domain as well as the snapshots. To memory massively parallel computer. A straight port,
facilitate this, the FFTs used in the pseudo analytical versus re-engineering of the code.
extrapolator (Etgen and Brandsberg-Dahl, 2009) were
replaced with distributed versions that run across all threads Having successfully ported RTM to a shared memory
in the node pool allocated for each shot. At this stage, this architecture and obtained better relative performance
was performed at math library level, switching out the compared to the one shot per node reference, the next step
standard Intel MKL FFT with the distributed MKL FFT. As was to add functionality to output azimuth sectored angle
can be seen in Figure 3, this helped performance, reducing gathers from the RTM (Frolov et al., 2016). This is further
the runtime by about 40% relative to the version relying on challenging the classical cluster compute model, as the
the parallel file system. However, this was still not output space is now increasing from a typical 3D stack
sufficient to get performance on par with running one shot image volume to a 4D or 5D volume, depending on how
per node.
Figure 4: An example of azimuth sectored angle gathers from RTM, left, with RMO tomography picks overlayed. The center
image shows angle gathers from a Gulf of Mexico dataset, where the angle domain is used to combine data from two separate
datasets. The right portion shows an example of how the sub-surface azimuth direction can be very useful for image
© 2017 SEG
optimization. Page 5275
many angular dimensions are used for the output image. An computer to be truly useful, it must have an acceptable
example of an azimuth sectored angle gather from RTM is programming model. If a model can offer better flexibility
shown in Figure 6 along with examples of how the angle and reduced turnaround in software development from
domain can be leverage for anything from velocity model concept to at-scale implementation, it might help offset the
building to advanced image optimization. cost of the pure hardware components when evaluated with
After successfully moving RTM to run on a shared memory a view to cost performance and cost of ownership. The use
computer, performance was retained, and new abilities of FPGSs comes to mind as a good example, notoriously
introduced. When dealing with very large, densely sampled hard to program, but with great performance. Their
input datasets and also wanting to keep data redundancy popularity in our industry has been on and off for quite
(pre-stack information) on output, a just as important some time, and we will for sure see them again in the near
challenge becomes memory size, bandwidth and overall future, but hopefully with a much better programming
system I/O performance. For the Gulf of Mexico FAZ model as part of the package.
survey described by Frolov et al. (2016), each output Maybe the key item that will help set the direction for the
location has 276 angle traces, so the RTM output volumes future is how fast any hardware/software combination can
for the about 12,000km2 survey was 276 times larger than a get an organization from concept to at-scale application.
stacked image. Data management and I/O will maybe be as Obviously cost performance will be a factor, but there
important as availability of pure compute Flops when it seems to be a trend towards an ever more rapid
comes to addressing the seismic imaging applications of the development cycle for imaging technologies, that then need
future. to be run on a large computer system to be applicable to the
With a need for more refined data decompositions and datasets from modern seismic surveys. People resources
analysis of data in multiple domains, it should be a safe bet enters this picture as well, so a holistic view on everything
to expect that problem sizes will grow. Add to this the from programming language, compilers and libraries, to
trends towards more use of inversion algorithms, the era of hardware configuration and power consumption will have
embarrassingly parallel computing might soon be behind to be addressed in any successful solution.
us.
Programmability versus performance; the future of Acknowledgement

HPC in the seismic industry
There is a large team of people to thank at PGS for having
If we as an industry are to continue to follow the trend of made the move to share memory computing possible. Of a
compute consumption shown in Figure 1, there are too long list to print here, I would like to acknowledge Raj
challenges ahead. The growth shown here was mainly Gautam, Sean Crawley and Pat Manning. Thanks to Cray
driven by a few key trends in HPC that the seismic industry Inc. for support throughout the process, and finally thanks
was able to follow and exploit fully. However, there is a to PGS for support and permission to present this work.
risk today that seismic imaging industry and HPC might
diverge. On one hand, the true mega compute centers of
today are server farms run by the Internet giants like
Amazon, Google and Microsoft, and they tend to have
hardware profiles very different from what we are used to
in the seismic or Oil & Gas industries. But with the rapid
growth seen in cloud computing, it might just be a
necessity for us to jump onto the trend and adapt our
problems to the prevailing hardware trends and
programming models used there. On the other hand, it is
still perfectly plausible that the classical HPC vendors will
continue to innovate and provide hardware platforms that
will prove themselves cost competitive also in the future. I
believe there will be continue focus on cost and turnaround,
so whichever compute platform can offer an edge will
prevail.
An important component in creating such a competitive
edge comes from the programming model and software
stack associated with the hardware. This is particularly true
when one considers the growing heterogeneity in hardware
with mixes of CPU, GPU and other accelerators. For a
© 2017 SEG Page 5276

EDITED REFERENCES
REFERENCES
Claerbout, J., 1985, Imaging the Earth’s interior: Blackwell Science.
Bleistein, N., 1987, On the imaging of reflectors in the Earth: Geophysics, 52, 931–942,
http://dx.doi.org/10.1190/1.1442363.
© 2017 SEG Page 5277


Phase Decomposition As A DHI in Bright Spot Regimes A Gulf of Mexico Case Study

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Phase Decomposition As A DHI in Bright Spot Regimes A Gulf of Mexico Case Study

Uploaded by

Copyright:

Available Formats

Phase Decomposition as a DHI in Bright Spot Regimes: A Gulf of Mexico Case Study

Energy, John Castagna, University of Houston

Summary the amplitude/phase/frequency relationship. Following

© 2017 SEG Page 3976

© 2017 SEG Page 3977

evident that separating the amplitude response in its phase

© 2017 SEG Page 3978

© 2017 SEG Page 3979

© 2017 SEG Page 3980

© 2017 SEG Page 3981

Adding the reflected P-waves to the direct waves (Fig- 4.55

-0.6 0.25 0.3 0.35 0.4 0.45 0.5

Woodford (the red dots in Figure 5) turns out to be instru-

in that plane (shown with the black, blue, and magenta

the direct waves only are not confined to the Woodford

Discussion and conclusions

reflector depth and dip poses immediate challenge, the

© 2017 SEG Page 3985

© 2017 SEG Page 3986

quality for better interpretation, fault definition, and sweet

NEAR-SURFACE MODEL AND DE-MULTIPLES

An extensive effort was made in de-multiple which

The study area is located in the Delaware Basin (Figure 1),

© 2017 SEG Page 3987

The shot gathers before (left) and after (right) de-noising

Figure 3. Autocorrelation before (left) and after de-multiple (right)

Figure 5. The shot before (left) and after de-noising (right)

Figure 4. Stack section after tomo-refraction statics (a) followed by

ADAPTIVE NOISE REMOVAL

De-noising of the data is a challenging problem since the

A two-pass approach was used for de-noising in the study

© 2017 SEG Page 3988

5D INTERPOLATION POST-STACK ENHANCEMENT

Figure 8 shows a cdp gather with NMO before (left) and

Figure 9 shows the stack section before (left) and after

Figure 9. Stack before (left) and after (right) interpolation.

The legacy data, the PSTM stacks previously processed by

© 2017 SEG Page 3989

Figure 13 shows a time slice of PSTM stack at 2000ms. CONCLUSIONS

The reprocessing has improved seismic images for

We thank Anadarko Petroleum Corporation for the

© 2017 SEG Page 3990

© 2017 SEG Page 3991

Summary greatly simplifies our analysis in comparison to other, more

© 2017 SEG Page 3992

Vint-slow values (red glyph colors). We interpret these

Interpretation areas that exhibit a high degree of natural fracturing

© 2017 SEG Page 3993

corresponding to highly fractured areas is consistent with

In addition to the strong visual correlation of cumulative

Figure 3: a) Cumulative gas production grid generated from

© 2017 SEG Page 3994

Once the multivariate analytics model was generated, we

Seismic inversion and VVAZ attributes were generated

© 2017 SEG Page 3995

© 2017 SEG Page 3996

The Permian Basin lies in West Texas and South Eastern

Figure 2: A 2007 time migration stack from the Delaware Basin

© 2017 SEG Page 3997

base of salt (BOS) image with very little distortion (Figure

© 2017 SEG Page 3998

reduction denoise (Sternfels et al., 2015) and 5D

improved noise attenuation coupled with the quadrupled