Geophysical Journal International

Geophys. J. Int. (2014) 197, 542–556 doi: 10.1093/gji/ggt526

Advance Access publication 2014 January 24

Utilizing the onset of time-lapse changes: a robust basis for reservoir

monitoring and characterization

D. W. Vasco,1 Thomas M. Daley1 and Andrey Bakulin2

1 Lawrence Berkeley National Laboratory, University of California, Berkeley, CA 94720, USA. E-mail: [email protected]
2 EXPEC Advanced Research Center, Saudi Aramco, Dhahran 31311, Saudi Arabia

Accepted 2013 December 29. Received 2013 December 27; in original form 2013 July 30

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

SUMMARY
Time-lapse monitoring is useful for imaging changes in geophysical attributes due to fluid
flow. In trying to use time-lapse data for reservoir characterization, difficulties often arise
when relating changes in geophysical observations to changes in fluid saturation and pressure.
As an alternative approach for reservoir monitoring and characterization, we introduce the
idea of an onset time, the time at which a measured quantity, such as a seismic traveltime
or a reflection amplitude, begins to deviate from its background value. We illustrate the idea
and demonstrate its utility through the consideration of traveltimes recorded by the continuous
active source seismic monitoring system at the Frio pilot site near Houston, Texas. The system,
which transmits an elastic wave every 15 min, is used to monitor the movement of carbon
dioxide injected into a permeable sand formation. From these data we can estimate the onset of
changes in seismic traveltimes to six receivers in an adjacent borehole. Numerical simulation
and synthetic tests indicate that the onset times are not very sensitive to the method used to
compute the effective fluid bulk modulus and, correspondingly, the seismic compressional
velocity. Rather, the onset times are strongly influenced by saturation changes within the
formation, specifically by the break-through time of the injected fluid phase. By means of
an iterative inversion algorithm we use the onset times to estimate permeability variations
between the boreholes at the Frio pilot site.
GJI Seismology

Key words: Inverse theory; Hydrogeophysics; Hydrology; Permeability and porosity;

Seismic tomography.

Various geodetic data sets, from borehole emplaced tiltmeters

1 I N T RO D U C T I O N
to satellite-based techniques such as the global positioning sys-
We are witnessing an increase in the use of time-lapse geophysical tem (GPS) and interferometric synthetic aperture radar (InSAR),
data for the monitoring of fluid flow and for the characterization of have been used to monitor fluid flow associated with environmen-
flow properties in the subsurface. This increase is observed across tal remediation, groundwater withdrawal, gas storage and energy
a wide range of geophysical subdisciplines. For example, there are production (Evans et al. 1982; Palmer 1990; Castillo et al. 1997;
a number of techniques from electrical and potential field time- Fielding et al. 1998; Wright 1998; Wright et al. 1998; Amelung
lapse methods that are used to monitor tracer migration, water table et al. 1999; Mossop & Segall 1999; Vasco et al. 2000; Du & Olson
movement and the tracking of plumes (LaBrecque & Yang 2001; 2001; Stancliffe & van der Kooij 2001; Vasco et al. 2002; Schmidt
McKenna et al. 2001; Day-Lewis et al. 2002; Miller et al. 2008; & Burgmann 2003; Oppliger et al. 2006; Vasco et al. 2010; Teatini
Jardani et al. 2009; Doetsch et al. 2010a; Martinez-Pagan et al. et al. 2011; Rucci et al. 2013). Even microseismic activity, associ-
2010; Karaoulis et al. 2011; Coscia et al. 2011, 2012). Time-lapse ated with the injection of fluids has been used in an effort to track
seismic techniques have been used in many contexts, from near sur- fluid migration (Shapiro et al. 1999). Finally, there have been at-
face environmental applications to deeper oil and gas and geother- tempts to combine multiple geophysical data sets to infer flow in
mal reservoirs (Greaves & Fulp 1987; Eastwood et al. 1994; Lee the subsurface (Hoversten et al. 2003).
et al. 1995; Mathisen et al. 1995; Watts et al. 1996; Lazaratos & Recently, there have been several studies extending the use of
Marion 1997; Wang et al. 1998; Tura & Lumley 1998; Johnston time-lapse data from simply monitoring fluid flow to subsurface
et al. 1998; Burkhart et al. 2000; Landro 2001; Smith et al. 2001; characterization. That is, to use time-lapse geophysical observations
Behrens et al. 2002; Guilbot & Brackin 2002; Arts et al. 2004; to estimate flow properties at depth. The level of sophistication
Osdal et al. 2006; Shi et al. 2007; Hodgson et al. 2007; Daley et al. varies from simple assisted history-matching of time-lapse data to
2007; Daley et al. 2008; Chadwick et al. 2012; Zhang et al. 2012). the fully coupled inversions of multiple data sets. Initial studies used
Published by Oxford University Press on behalf of The Royal Astronomical Society 2014. This work is written by (a) US Government
542 employee(s) and is in the public domain in the US.
 Utilizing the onset of time-lapse changes 543

time-lapse seismic data to estimate reservoir permeability variations This work follows on from two previous studies in which geodetic
and to identify fast flow paths and barriers to flow (Landa & Horne data were used to infer the propagation of volume change within a
1997; Huang et al. 1998; He et al. 1998; Gosselin et al. 2003; Vasco reservoir (Vasco 2004b, Rucci et al. 2010). Vasco (2004b) demon-
2004a, Vasco et al. 2004; MacBeth & Al-Maskeri 2006; Daley et al. strated that the onset of the most rapidly variation in reservoir
2011). Kowalsky et al. (2004, 2006), Camporese et al. (2011) and volume was directly related to the spatial variation of the flow prop-
Zhang et al. (2014) have used time-lapse electromagnetic data, erties within the reservoir. Rucci et al. (2010) indicated that this
such as ground penetrating radar and resistivity tomography, to relationship was not sensitive to the spatial variations of geome-
determine flow properties at depth. Geodetic data have also been chanical properties within the reservoir. The theoretical result was
used for characterization (Vasco et al. 2001; Vasco 2004b, Vasco verified in an application to InSAR data from a carbon capture and
& Ferretti 2005; Rucci et al. 2010). Such data are typically very storage project at In Salah, Algeria. In this paper, we formalize the
cost-effective, allowing for dense temporal sampling. The variation approach, based upon the notion of an onset time given above, and
in microseismicity has been used to infer the hydraulic diffusivity illustrate its application to a time-lapse crosswell seismic data set.
at depth (Shapiro et al. 1999; Shapiro et al. 2002). As in the geodetic applications, we find that a collection of onset
While time-lapse observations have proven useful for reservoir times, derived from the crosswell data, provide a robust data set with

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

characterization, a fundamental challenge remains. The difficulty which we may infer flow properties. In particular, the relationship
is in specifying the relationship between saturation and pressure between onset times and flow properties, is not sensitive to the rock
changes within a reservoir and variations in geophysical properties physics model. One may take advantage of this characteristic in a
such as seismic velocity, electrical conductivity and volume strain. wide array of applications.
Consider the relationship between seismic compressional velocity
and fluid saturation changes. It is well known that this relationship
is complicated by a host of factors, such as the spatial variation of 2 METHODOLOGY
rock properties and/or fluid distribution at a scale larger then a core In this section, we relate the onset time to the flow properties of
sample but smaller then a seismic wavelength, the mesoscopic scale the porous medium, such as the porosity or permeability. As a
(White 1975; Pride et al. 2004). Determining the mesoscopic vari- specific illustration, we consider seismic time-lapse monitoring of
ations in rock properties and fluid saturations in a reliable fashion carbon dioxide injected into a heterogeneous reservoir. In truth, the
is an extremely difficult task. Thus, such information is unlikely to methodology is applicable to any time-lapse data set with a temporal
be available for most characterization efforts. Furthermore, there sampling that is dense enough to capture the onset of a change in a
are other factors that can confound the situation, such as physi- geophysical attribute.
cal and chemical changes within the rock caused by the invading
fluid (Ghosh & Sen 2012). Such factors can be time-dependent and
are not currently well understood. Given the significant variability 2.1 Modelling flow: relating saturation changes to the flow
in rock types within the Earth, it may prove difficult to develop properties of the porous medium
a sufficiently general model with properties that can be reliably
determined by geophysical observations. 2.1.1 Governing equations for saturation and pressure changes
Here, we take a different approach, based upon the idea of an
Consider a porous medium containing two fluid phases. We denote
onset time. An onset time is that point in time at which a geo-
the saturation of the aqueous phase by Sw (x, t), while the saturation
physical property begins to deviate from its background, or initial,
of the non-aqueous phase is represented by Sn (x, t). Both saturations
value. The estimation of onset times requires sufficient temporal
vary as functions of the spatial x and temporal t coordinates. Since
sampling to capture the initiation of a change in a geophysical
the pore space is completely filled by the fluids, the saturations must
observation. In many situations, the onset time can be related to
satisfy the relationship
saturation and/or pressure changes within the reservoir. In the ex-
ample presented below, onset times are associated with the arrival Sw + Sn = 1. (1)
of injected carbon dioxide at various locations in the subsurface.
Rather than invert the seismic traveltime changes themselves, we The pressure associated with each fluid within the pore space may
invert the onset time of the changes. Such an approach removes differ, introducing surface tension on the interface between the fluids
the requirement that we map a change in a geophysical property and causing one fluid to impinge upon the other. This difference,
to a specific change in fluid saturation and/or pressure. Rather, we the capillary pressure (pc ), is given by
must relate the initiation of a change in the geophysical property to pc = pn − pw , (2)
the arrival time of a saturation front and/or the time at which the
fluid pressure changes most rapidly. Such fluid phase arrival times where pn is the pressure in the non-wetting or non-aqueous phase
can often be linked to flow properties. For example, as shown in and pw is the pressure in the wetting phase. It is generally assumed,
Vasco & Datta-Gupta (1999) and Vasco & Finsterle (2004), tracer based upon empirical evidence provided by laboratory experiments,
arrival times have a rather direct relationship to spatial variations that the capillary pressure is a function of the saturation (Peaceman
in permeability. Similarly, it was noted in Vasco et al. (2000) the 1977, p. 14):
arrival time of the largest fluid pressure change has a much simpler pn − pw = pc (Sw ). (3)
relationship to flow properties then does the magnitude of the fluid
pressure changes. Due to the non-linearity associated with mul- Fluid velocities are given by two-phase generalizations of Darcy’s
tiphase flow, the arrival time of a saturation change is somewhat law, which states that the velocities for the aqueous and non-aqueous
more complicated. In a series of papers (Vasco et al. 1999; Vasco & phases are driven by their respective pressure gradients and gravi-
Datta-Gupta 2001; Vasco et al. 2004; Vasco 2011) have shown that tational forces
the saturation arrival time has a direct relationship to reservoir flow K kr w
vw = − (∇ pw − ρw gZ) , (4)
properties. φμw
544 D.W. Vasco, T.M. Daley and A. Bakulin

K kr n
vn = − (∇ pn − ρn gZ) , (5)
φμn
where K(x) is the absolute permeability which depends upon spatial
position x but not upon the time, φ(x) is the porosity, μw and μn
are the fluid viscosities, ρ w and ρ n are the fluid densities, g is
the gravitational acceleration, Z is a unit vector in the direction
of the gravitational field and krw (Sw ) and krn (Sw ) are the relative
permeabilities, given by the ratio of the effective permeability of
each fluid to the absolute permeability K (Peaceman 1977, p. 15):
Kw
kr w (Sw ) = ≤ 1, (6)
K
Kn
kr n (Sw ) = ≤ 1. (7)
K

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

By combining the equations for the conservation of each phase
and the two-phase generalization of Darcy’s law, one may derive
the coupled equations governing two-phase flow in a heterogeneous
porous medium (Peaceman 1977, p. 16):
 
ρn K kr n ∂ (φρn Sn )
∇· (∇ pn − ρn gZ) + qn = , (8)
μn ∂t
 
ρw K k r w ∂ (φρw Sw )
∇· (∇ pw − ρw gZ) + qw = , (9) Figure 2. Reference permeability model used to generate synthetic travel-
μw ∂t time changes induced by the injection of carbon dioxide into a central well,
denoted by the unfilled circle. The variations in tone denote changes in the
where qn and qw are the source or sink terms for the respective
logarithm of permeability. The monitoring well in indicated by the unfilled
phases. Eqs (8) and (9), being non-linear and containing spatially star 30 m to the north of the injector. The straight line connecting the two
varying coefficients, are typically solved numerically. Here, we use wells signifies the intersection of a crosswell plane with the reservoir layer.
the numerical reservoir simulator TOUGH2 (Pruess et al. 1999;
Pruess 2004) to solve the coupled system of eqs (8) and (9) for a 2.1.2 An equation for saturation arrival times
particular reservoir model and specific boundary conditions. For our
By solving the governing equations numerically we can derive sat-
synthetic application we simulated the injection of carbon dioxide
uration and pressure histories for each grid block in our reservoir
into a 2.5-m thick layer. Variably-sized grid blocks are used to
model. We can construct the saturation arrival time from these his-
capture the rapid changes in fluid saturations and pressures near the
tories and the corresponding propagation time from the injection
wells (Fig. 1). A heterogeneous permeability model was generated
well to each grid block. In this way, results from the numerical sim-
by spatially smoothing a random field (Fig. 2).
ulation can be used to construct a path X(xo , t), specified by the
observation point xo and the time t, extending from the injector to
the observation point. One simply follows the saturation evolution
backward in time, starting from the arrival time at a given observa-
tion point back to the beginning of injection at the well. At any point
along that path the instantaneous velocities of the phases are given
by eqs (4) and (5). From the path X and the instantaneous fluid
phase velocities we can calculate the arrival time of the injected
fluid and the traveltime along the path. For example, in the case of
the non-aqueous phase, the traveltime is given by

dr
τ= (10)
X |vn|

or, using eq. (5),


φμn
τ= dr (11)
X K kr n |Un |

where
Un = ∇ pn − ρn gZ. (12)
Expression (11) provides a relationship between the arrival time
of the non-aqueous phase and the flow properties of the porous
medium. As such, it will serve as the basis of our inversion algo-
rithm. Note that the velocity vector vn contains both an explicit
Figure 1. Geometry of the grid used in the numerical simulation of the and an implicit dependence upon the permeability. The implicit de-
injection of carbon dioxide at the Frio pilot site. Pressure variations are pendence follows from the fact that the expression for Un contains
indicated by the small coloured squares at the center of each grid block. the pressure gradient ∇pn which is a function of the permeability
 Utilizing the onset of time-lapse changes 545

distribution within the reservoir. As noted in the Appendix [eq. onset times and we must solve for an effective permeability change,
(A10), also see eqs (1–57) in Peaceman (1977)], the velocity vector δK = δK − δφ. From this point on we shall drop the prime and
vn also depends upon the fractional flow properties of the two fluids treat this quantity as the effective permeability of the medium. It
and the capillary pressure curve. For brevity, we shall not discuss might be possible to incorporate other types of data to resolve this
this in further detail. We will assume that the overall relative per- trade-off. Under these restrictions, eq. (14) takes the form
meability and capillary pressure properties of the formations are  
δK 1 δK
characterized by laboratory experiments. Furthermore, we shall re- δτ = − I dr = − dr. (16)
X K X |vn| K
compute the vector vn at each iteration of our inversion algorithm
when we compute the sensitivities. An alternative is to simply solve for the ratio φ/K, which is better
constrained by the onset time data.

2.1.3 Sensitivities: relating perturbations in permeability to

perturbations in saturation arrival times 2.2 Relating changes in seismic arrival times to saturation
For our iterative inversion algorithm we adopt the general approach changes and onset times

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

described in Vasco et al. (2004), in which reservoir permeabili- Our observations will be seismic traveltimes for elastic waves gen-
ties are updated in order to match the onset times. We begin our erated by a source in one borehole and recorded by a set of hy-
derivation by writing the integrand in eqs (10) and (11) as drophones in an adjacent borehole, as described in Daley et al.
1 φμn (2011). The overall set-up (Fig. 3) consists of a single source in the
I (r, t) = = , (13) injection well. The source is situated somewhat above the injection
|vn | K kr n |Un |
interval and activated every 15 min. The transmitted elastic waves
where we have included a dependence upon t because the vector field are recorded in six borehole receivers in the observation well. The
vn can change with time, particularly if there is a change in injection background elastic velocity structure was derived from a sonic log
or production. A perturbation in the integrand due to a change in in the injection well at the actual Frio site. The boundaries of the
the model parameters leads to a perturbation in the arrival times reservoir layer, which is 2.5 m thick and dips 18◦ , are indicated by
according to: the two solid lines in Fig. 3.

Currently, theories relating changes in fluid saturation to changes
δτ = δ I (r, t) dr. (14)
X
in the propagation velocity of elastic waves in a porous medium
The perturbation in flow properties will generally change the flow
field and hence the integration path X. As in geophysical tomo-
graphic algorithms, we have ignored this effect and integrated along
the path computed using the background reservoir model.
Assuming that the fluid, the relative permeability and the capil-
lary pressure properties are fixed, perturbations may occur in the
porosity, the permeability, and because of its implicit dependence
on porosity and permeability, the velocity vector Un may also un-
dergo a perturbation. A general perturbation in I(r, t) may therefore
contain the following contributions:
∂I ∂I ∂I
δ I (r, t) = δφ + δK + δ|Un |. (15)
∂φ ∂K ∂|Un |
The partial derivatives are obtained from the analytic form for I(r, t),
eq. (13). As in Vasco et al. (2004), we shall neglect the implicit sen-
sitivity of Un to changes in the porosities and permeabilities in a
single grid block of the reservoir model. In computing sensitivi-
ties to a change in a single grid block, we shall assume that the
effect of a perturbation to the flow field, vn , does not change the
perturbation in the saturation traveltime. Numerical tests in Vasco
& Datta-Gupta (2001) and Vasco et al. (2004) indicate that this is
an adequate approximation. Thus, we neglect the term associated
with perturbation in δ|Un |.
Vasco et al. (1999, 2004) noted the tradeoff between changes in
porosity and changes in permeability. That is, the porosity sensitivity
is the negative image of the permeability sensitivity:
∂I ∂I
=−
∂φ ∂K
and one may write eq. (15) as
∂I ∂I Figure 3. Compressional wave velocity variations between the injection
δ I (r, t) = (δ K − δφ) = δ K , (left edge) and observation (right edge) boreholes. The unfilled star indicates
∂K ∂K the location of the seismic source. The open circles denote the locations of
where we have neglected the δ|Un | term, as noted above. Therefore, the labelled receivers. Seismic raypaths are indicated by the trajectories
it is not possible to resolve these parameters independently using connecting the source to the receivers.
546 D.W. Vasco, T.M. Daley and A. Bakulin

are active areas of research (Müller et al. 2010). It appears that the Voigt-Reuss bounds are valid in the presence of anisotropy while
relationship between a variation in fluid saturation and a change in the Hashin-Shtrikman bounds are only valid for a macroscopically
elastic velocity depends upon the distribution of the fluids within homogeneous and isotropic material.
the pore space at a scale smaller then that of the seismic wavelength The effective bulk modulus of the entire saturated rock, Ks , in-
(White 1975; Johnson 2001). In order to explain the variation in cluding the two fluids (water and carbon dioxide), the properties of
seismic wave speed and attenuation as a function of frequency, the rock constituents or grains, and the frame moduli is given by
it is necessary to invoke variations in medium properties or fluid Gassmanns (1951) relation
saturations on a length scale smaller then the seismic wavelength
(1 − K d /K g )2
and larger then the pore scale, the so-called mesoscopic scale. In K s (x, t; τ ) = K d + , (20)
this model, the heterogeneity necessitates fluid pressure equilibra- φ/K f (x, t; τ ) + (1 − φ)/K g + K d /K g2
tion over a time interval that is related to the spatial length-scale where Kd is the dry frame bulk modulus, Kg is the bulk modulus of
and the variations in properties. This notion, recently advocated by the solid grains comprising the rock, Kf is the fluid bulk modulus
Pride et al. (2004) in a generalization of earlier studies, has been given by the Voigt (17), Reuss (18), or Hill (19) prescriptions. The
elaborated and utilized by others. For an overview of many related density of the saturated rock is given by

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

contributions, see the review by Müller et al. (2010). The implica-
tion of this mesoscale structural dependence on estimates of onset ρs (x, t; τ ) = φρ f (x, t; τ ) + (1 − φ) ρg (21)
times is not altogether clear and is explored next. Specifically, in the where
next subsection we consider upper and lower bounds on seismic ve-
locity changes. Following that, we examine how estimates of onset 
N
ρ f (x, t; τ ) = Si (x, t; τ )ρi , (22)
times vary as these bounds are attained. i=1

and ρ i is the density of the ith fluid under reservoir conditions.

2.2.1 Relating changes in fluid saturation to changes We assume that the shear modulus is not strongly influenced by
in seismic velocity the presence of the fluids and may be approximated by the shear
modulus of the unsaturated rock, Gd . The seismic compressional
The specific upper and lower limits that we shall consider are the
velocity, Vp , is given by the expression
classic Voigt and Reuss bounds (Mavko et al. 1998). The Voigt

upper bound for the bulk moduli of a mixture of N fluids is given K s (x, t; τ ) + 43 G d
by V p (x, t; τ ) = (23)
ρs (x, t; τ )

N
K v (x, t; τ ) = Si (x, t; τ )K i , (17) for a given saturation. The Voigt-Reuss velocity bounds are plotted
i=1 in Fig. 4 as functions of the saturation of carbon dioxide, along with
the estimates of Hill (1963). These bounds were computed using
where Si (x, t; τ ) is the saturation of the ith fluid and Ki is the bulk
the properties associated with the Frio sandstone formation and the
modulus of the ith fluid. We use a notation in which the dependence
pressure and temperature conditions of the reservoir. The difference
upon the arrival time of the saturation change, τ , given by eq. (10), is
between the upper and lower velocity values is significant, of the
explicit. That is, the saturation begins to change when t exceeds τ at
order of the entire velocity variation due to changes in the fluid
a particular location. The magnitude of the saturation change from
saturation. Such large differences highlight the importance of the
that time on depends upon t − τ . The Voigt upper bound is obtained
when all the contributing components are arranged in parallel. One
can visualize this as layers of pure components arranged parallel to
the direction of compression. Then the stiffest layer determines the
compressibility and controls the modulus. The Reuss lower bound
for the effective fluid bulk modulus is given by
 N −1
 Si (x, t; τ )
K r (x, t; τ ) = . (18)
i=1
Ki

This lower bound is obtained when all the components are arranged
in series. The Reuss bound can be thought of as compression in
a direction perpendicular to layers of pure components. Thus, the
weakest layer plays a significant role in the compressibility of the
stack of layers and, to a large degree, determines the modulus. The
Hill average fluid bulk modulus (Hill 1963) is simply the average
of the Voigt and Reuss Bounds:
K v + Kr
K h (x, t; τ ) = . (19)
2
Bounds such as these are used in a variety of physical settings, as in
thermal conductivity where they are known as the Wiener bounds
(Wiener 1910; Tong et al. 2009). Other bounds, such as the Hashin- Figure 4. Velocity bounds computed using the Voigt [eq. (17)] and Reuss
Shtrikman bounds (Hashin & Shtrikman 1963), are possible but [eq. (18)] algorithms. The Hill velocity estimate, which is the average of
the Voigt-Reuss bounds are the simplest and the most conserva- the two bounds [eq. (19)] is also plotted as a function of the carbon dioxide
tive for a particular volumetric mixture of fluids. For example, the saturation.
 Utilizing the onset of time-lapse changes 547

rock physics model when interpreting the magnitude of seismic the ray path, the velocity is lowered and the traveltime increases.
velocity and amplitude changes (Mavko et al. 1998). Note that if the saturation change were associated with an increase
in velocity then it may be possible for the ray path to shift before the
fluid front arrives. For the injection of carbon dioxide the velocity
2.2.2 Computing seismic traveltimes will decrease and the ray will tend to bend around the zone of lower
The basic observations are seismic traveltimes, T(x, t; τ ), the times velocity. Our modelling indicates that the shifting of the ray paths
required for the propagation of an elastic wave from the source to a is generally not significant up to the onset time, and it is acceptable
set of hydrophones in the observation well. Given a velocity model to use the background velocity model to compute the rays up to the
Vp (x, t; τ ), high-frequency seismic first arrival times are calculated onset time.
by solving the Eikonal eq. (Chapman 2004, p. 139)
1
∇T · ∇T = . (24) 2.2.3 The influence of the effective fluid bulk modulus on the
V p2 (x, t; τ )
calculated traveltime changes and onset times
The surface
As an illustration, consider the crosswell geometry shown in Fig. 5.

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

T (x, t; τ ) = C (25) The dipping layered velocity structure (Fig. 3) serves as the back-
for some constant C defines a wave front, a surface of constant ground model. Saturation changes lead to velocity decreases within
seismic traveltime from the source. As indicated in Chapman (2004, the reservoir layer, as indicated by the panel on the right. The mag-
p. 139) one can consider a ray path R(s), the trajectory orthogonal to nitude of the velocity decrease, calculated using Hill’s method,
the wave fronts, where s is the distance along the path. In coordinates 0.5 km s−1 , agree with the P-velocity changes seen in the Frio-I
oriented along the ray, the eikonal equation reduces to pilot time-lapse crosswell experiment (Daley et al. 2008). To com-
pute the ray paths we solve the Eikonal eq. (24) numerically using
dT 1
= . (26) a finite-difference algorithm as proposed by Vidale (1988) and im-
ds V p (x, t; τ ) plemented by Podvin & LeComte (1991). The resulting traveltime
By integrating (26) along the ray path we arrive at an expression for fronts, corresponding to 5 d of injection, are shown in the left-hand
the traveltime of a compressional elastic wave panel of Fig. 5. The seismic ray paths are obtained using an iterative
 technique to march down the gradient of the travel field from the
ds
T (x, t; τ ) = . (27) receivers in the direction of decreasing traveltime. The important
R V p (x, t; τ ) component of the ray path is that part intersecting the reservoir layer.
This expression provides a mapping between saturations and seismic The first encounter with a saturation change will lead to the onset of
traveltimes. We can use it to relate the arrival of the injected fluid a change in the arrival time of the seismic wave at the receiver. We
phase at points along the ray path to the onset of changes in the define the onset time as the calendar time at which the traveltime
traveltime of the seismic wave. The onset time for a particular change exceeds 10 per cent of the peak change. This value is arbi-
receiver is due to saturation changes encountered by the segment of trary but is thought to represent a reasonable noise level associated
the ray path within the reservoir. When the saturation front crosses with time-lapse estimates. The exact value used in applications will

Figure 5. (Left) Contours of traveltime for an elastic wave generated at the source (unfilled star). The contour interval is 1 ms in this plot. The background
velocity model for the traveltime computations is plotted in Fig. 3. (Right) Velocity deviations from the background velocity model due to the injection of
carbon dioxide into the reservoir layer. The raypaths are indicated by the trajectories from the seismic source to the six receiver locations (open circles).
548 D.W. Vasco, T.M. Daley and A. Bakulin

flow model (Fig. 1). When the carbon dioxide enters a grid block
the saturation in the entire grid block is changed by the simulator.
Thus, the saturation of entire 3-m regions change abruptly in our
reservoir simulation, leading to corresponding jumps in seismic
velocity. A finer discretization would reduce these rapid changes,
with a corresponding increase in computation time. The changes
in traveltime for the four deepest receivers (1666, 1670, 1676 and
1680 m) occur first, at around 0.1 to 0.3 d. The early onset of a
traveltime change at the deepest receivers is in accordance with
our intuition, because these four ray paths intersect the reservoir
layer nearest to the injector, some 10 to 15 m from the injection
well (Fig. 5). For the receiver located at a depth of 1658 m in the
observation well, the initial change in traveltime does not occur until
about 1 d after the start of injection. The ray path for this receiver

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

intersects the reservoir layer somewhere between 15 and 25 m from
the injection well. The first notable changes in the traveltime to
the shallowest receiver (1654 m) do not occur until around 4.5 d
following the beginning of injection. The later onset time agrees
with the fact that this raypath intersects the reservoir layer furthest
from the injection site, at a distance of 30 m or more. Note also, that
the magnitude of this traveltime change, less than 0.1 ms, is much
Figure 6. Deviations from pre-injection traveltimes as a function of elapsed
time since the start of injection. The traveltimes were calculated from the smaller than the 0.3 to 0.4 ms changes found at the other receivers.
reservoir simulations using the Hill algorithm (19). The onset times are An explanation for this smaller change in arrival time is that the ray
indicated by the unfilled squares. path for this receiver has a much shorter component within the high
permeability layer and thus encounters very little carbon dioxide
depend upon the characteristics of the particular field site and the along its travel path (Fig. 5).
experimental setup. For example, we can use the pre-injection time Now consider how variations in the method used to compute
series to estimate the noise-level. the effective fluid bulk modulus influences the magnitude of the
The changes in seismic traveltime for each of the six receivers are traveltime change and the onset time at each receiver. We use the
plotted in Fig. 6 as a function of the elapsed time since the start of Reuss (18) and Voigt (17) algorithms to map saturation changes
injection. A value of zero in this plot indicates no deviation from the into moduli and velocity changes used in the eikonal equation. The
traveltime observed before the injection. For these calculations we calculated traveltime changes following the start of carbon dioxide
use the approach of Hill (1963) in which the Voigt and Reuss bounds injection are plotted in Fig. 7. There are significant variations in
are averaged [see eq. (19)]. The somewhat discontinuous changes the size of the traveltime changes, around a factor of five, between
in the traveltime deviation are due to the 3 m grid spacing in the the Voigt and Reuss values. Note that the time at which significant

Figure 7. Deviations from pre-injection traveltimes as a function of elapsed time since the start of injection. (Left) The traveltimes were calculated from the
reservoir simulations using the Reuss algorithm. (Right) The traveltimes were calculated from the reservoir simulations using the Voigt algorithm. The open
squares denote the onset times at each source–receiver pair.
 Utilizing the onset of time-lapse changes 549

is iterative in the sense that we start with a reservoir model and

modify it in a succession of updates designed to better match the
observed onset times. The onset times are derived from the seis-
mic traveltime histories. Given sufficient ray coverage, one could
conduct a sequence of seismic tomographic inversions to construct
velocity histories for the grid blocks of the reservoir model. Then
the onset time for each grid block may be derived from the velocity
histories. For the synthetic case treated here (Fig. 5) we only have
six observations for the seismic receivers. The onset times are com-
puted from the traveltime changes (Fig. 6). We attribute the onset
time to a saturation change in the reservoir grid blocks that the
seismic ray traverses. It is assumed that the carbon dioxide propa-
gates away from the injection point and that the propagation time of
the carbon dioxide increases monotonically from the injector. Thus,

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

we assume that earliest seismic traveltime change, in particular the
onset time of the change, is due to saturation changes in the grid
block of the reservoir model traversed by the ray path and nearest
to the injection well. We construct the flow path from the center
of this block back to the injector to obtain the path X used in the
integral (16).
The basis for the inversion algorithm is a discrete version of
eq. (16)

N
li j
τi = − Ij δK j, (28)
j=1
Kj

where τ i denotes the change in saturation arrival time (onset

time) for the ith observation, the sum is over the N grid blocks of
the reservoir model, Kj is the background permeability of the jth
grid block, δKj is the permeability update for that grid block, and
lij is the path length for the ith trajectory, Xi , in the jth grid block.
If the trajectory does not intersect a particular grid block then lij
is zero and there is no contribution to the sum. For a given set of
onset times, we will have a collection of equations, each in the form
of the linear constraint (28). We may write the system of equations
in matrix-vector form
d = AδK, (29)
where the elements of A are given by
li j 1 li j
Ai j = −I j =− , (30)
Figure 8. (Top) Peak traveltime change for each of the six receivers plotted Kj |vn | j K j
at their respective depths. The three curves signify the three methods (Voight,
d is the vector of onset time residuals and δK is the vector containing
Reuss, and Hill) for computing seismic velocities. (Bottom) Onset times
for each receiver plotted at the depth of the instrument in the observation
the permeability update. At every iteration we seek updates that
borehole. improve the match to the observed onset times or minimize the
magnitude of the squared misfit vector
changes are first observed at each receiver, the onset time, denoted (δK) = (d − AδK)T (d − AδK) , (31)
by the unfilled squares in Fig. 7, does not change significantly when
either the Reuss or Voigt approach is used. These observations are possibly subject to additional regularization terms penalizing at-
encapsulated in the two panels in Fig. 8. The upper panel displays tributes of the model such as its spatial roughness or its vector norm
the peak traveltime changes for the three approaches (Hill, Reuss (Menke 1989). The vector δK that minimizes (31) is obtained by
and Voigt) for all receivers. The factor of five variation in magnitude solving the necessary equations for a minimum (Noble & Daniel
is clear in this figure. The lower panel is a plot of the onset times 1977, p. 60)
for each set of estimates. The onset time does not vary significantly AT AδK = AT d. (32)
when the algorithm for computing the effective fluid bulk modulus
is changed. The variations are roughly 0.1 to 0.2 d, which is less than We solve the linear system (32) using the iterative LSQR algorithm
5 per cent of the peak value of the onset time, which is almost 5 d. of Paige & Saunders (1982). For the mth iteration, the update is
given by
Km = Km−1 + δKm , (33)
2.3 An iterative inversion algorithm for permeability
and for each iteration we conduct a full reservoir simulation using
We now outline an algorithm for estimating permeability variations the previous model Km − 1 to compute the elements of the matrix A
within a reservoir given a collection of onset times. The algorithm and the residuals.
550 D.W. Vasco, T.M. Daley and A. Bakulin

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

Figure 9. Mean squared error as a function of the number of iterations of Figure 10. Estimated permeability variations between the injection and
the inversion algorithm. The three curves signify inversions for the Hill, observation wells. Each curve is associated with a distinct inversion, based
Reuss, and Voigt onset times. upon Hill, Reuss, and Voigt onset times.

To examine the impact of varying the method used to calculate

the effective fluid bulk modulus, we consider inversions using the
Hill, Reuss, and Voigt onset times (Fig. 8). The Hill algorithm is
used to calculate the sensitivities and the residuals for each inver-
sion. The synthetic onset times d are computed using the three
approaches (Hill, Reuss, and Voigt). The resulting misfit reduction,
as a function of the number of iterations, is essentially identical
for the Hill and Reuss onset times (Fig. 9). For the Voigt onset
times the final misfit is marginally larger but the total reduction is
still almost two orders of magnitude and the final misfit is quite
small.
In order to evaluate the impact of the different rock physics mod-
els on the inversion of onset times, we compare the three sets of
permeability estimates, Voigt, Reuss, and Hill, with each other and
with the reference permeability variation between the two bore-
holes. In Fig. 10, we plot the variation of the permeability in the
reservoir layer between the wells for the inversions of the three
sets of onset times. The position of this line within the reference Figure 11. Reference and estimated interwell permeability variations. Two
model is plotted in Fig. 2. The agreement between the three esti- reference permeability variations, one directly along the line between the
mates is good and each model indicates increasing permeability as two wells and one offset 3 m to the west of the line between the wells, in a
the observation well is approached. This overall increase in per- higher permeability region.
meability from the injection well to the observation well is also
3 A P P L I C AT I O N : C O N T I N U O U S A C T I V E
seen in the reference model (Fig. 2). A more detailed compari-
S E I S M I C S O U RC E M O N I T O R I N G
son is presented in Fig. 11 in which we plot the Hill permeability
(CASSM) OF THE INJECTION OF
multiplier estimates along with those of the reference model. We
C A R B O N D I O X I D E AT T H E F R I O S I T E
plot two cross-sections through the reference model, one directly
on the line between the two boreholes and a parallel cross-section
3.1 Overview
offset 3 m to the west in a region of higher permeability. One
of the difficulties associated with interpretating crosswell observa- The Frio I and II brine pilot experiments involved the monitoring of
tions is the fact that fluid may flow out of the plane that is imaged carbon dioxide injected into deep (1664 to 1666 m), high porosity,
by the seismic data. In the present case higher permeabilities to sand formations. The dipping sand layers flank a salt dome in the
the west leads to significant fluid flow adjacent to the crosswell vicinity of Houston, Texas. Some 380 tons of carbon dioxide were
plane, leading to an apparent higher permeability. Thus, the perme- injected over a period of 5 d at an average rate of 0.9 kg s−1 .
ability estimates follow the highest permeability values of the two The migration of the injected carbon dioxide was monitored by
cross-sections. both crosswell seismic techniques and geochemical fluid sampling
 Utilizing the onset of time-lapse changes 551

(Freifeld et al. 2005; Daley et al. 2007). The arrival of the carbon
dioxide at the monitoring well, some 2 d after the start of injection,
was detected by continuous U-tube fluid sampling in the observation
well (Daley et al. 2011).
The small-scale pilots were useful for calibrating models and
techniques at intermediate length scales, somewhere between the
core and log scale and the scale provided by surface seismic obser-
vations (Daley et al. 2011). The target formation for the Frio II pilot
experiment was the 17 m thick fluvial Blue sand. The sediments in
the area dip 18◦ from the observation well towards the injection well
30 m away (Daley et al. 2011). Logs taken in the injection well in-
dicate variable permeabilities in the region of injection, with values
as high as 4 Darcies in isolated locations and average porosities of
25 to 30 per cent. Of particular importance for the experiment was

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

the existence of a 2–3 m sequence of generally high permeability
layers intersecting the injection interval between 1663 and 1666 m
in depth. Two permeability logs, obtained using distinct methods,
indicate the presence of this zone but displayed different internal
structure and also differ to some degree outside of this zone. The
indirect estimates provided by the permeability logs were calibrated
using core measurements. The resulting high permeabilities were Figure 12. Observed traveltime changes as a function of the time since the
later validated during pressure interference testing conducted in ad- start of injection. The six curves are associated with instruments at different
vance of the injection (Daley et al. 2011). A sonic log run in the depths in the observation well. The onset times are indicated by the open
injection well provided the background velocity model, which is squares.
identical to that used in the illustration presented earlier (Fig. 3).
There is fair agreement between the observed and calculated pre- The general features of the traveltime changes in Fig. 12 are
injection traveltimes at each of the six receivers. similar to those found in our numerical illustration (Fig. 6). In
particular, the deeper receivers (1666, 1670, 1676 and 1680 m)
cluster with early onset times while the shallower receivers (1658
and 1654 m) have successively larger onset time. The onset times
are much shorter for the actual Frio observations. We found that this
3.2 CASSM observations was a density effect, due to the dip of the formation and the fact that
While the U-tube sampler provides an estimate of the large-scale the carbon dioxide is less dense than the in situ saline water. This
fluid traveltime from the injector to the observation well, it does gravitational effect was not modelled in the numerical illustration
not reveal much about propagation of the carbon dioxide in the re- but was included in modelling the actual Frio II experiment. Both
gion between the wells. More details on the interwell propagation the predicted traveltime changes based upon the Hill model and the
are provided by the Continuous active-source seismic monitoring observed traveltime changes (Fig. 12) increase rapidly around the
(CASSM) system, designed and deployed at the Frio site by sci- onset time and quickly plateau to constant values. Finally, the mag-
entists from Lawrence Berkeley National Laboratory (Daley et al. nitude of the traveltime change is much smaller at the shallowest
2008; Daley et al. 2011). The overall CASSM source-receiver ge- receiver (1654 m). This is most likely due to the fact that receiver is
ometry is indicated in Fig. 5. A single source, located above the situated near the top of the formation and the seismic energy does
formation, is used to propagate elastic waves to receivers in an not propagate through the entire carbon dioxide saturated layer
adjacent borehole. Twenty-four hydrophone sensors were deployed (see Fig. 5). A notable difference between the synthetic traveltime
but only 13 provided usable data due to electrical leakage. Of the changes and the observed changes is the smaller magnitude of the
13 instruments, we use six which recorded elastic energy prop- observed traveltime change associated with the receiver at a depth
agating through the formation and with sufficient signal-to-noise of 1658 m. This may be due to a variation in the properties of the
for the estimation of onset times (Fig. 12). The receivers above rock physics model mapping from saturation to seismic velocity.
the high-permeability layer were not used to avoid the effects of The sensitivity to the rock physics model may be seen by compar-
possible density-driven vertical migration of carbon dioxide within ing the relative magnitudes of the Reuss and Voigt models in Fig. 7.
the well itself. In particular Daley et al. (2011) note that one hy- Alternatively, it might be due to a thinning of the formation near
drophone, at a depth of 1650 m, had a significantly larger trav- the observation well or some other deviation from our reservoir
eltime increase of over 1 ms. The waveform associated with this modelling.
hydrophone changed dramatically during the injection, leading to
difficulties in consistently identifying the onset time. Therefore, un-
3.3 Onset times and estimates of permeability variations
til we can better resolve the situation for the shallow hydrophones,
we have decided to focus on receivers at or below the depth of the From the traveltime histories in Fig. 12 we estimated the onset times,
high-permeability layer. Due to interference with a downhole pres- the calendar time at which the traveltime change exceeds 5 per cent
sure gauge the recording was delayed until 2 hr after the start of of the peak change. We used a smaller value to compute the onset
injection. During its operation four pulses per second were recorded time because the signal-to-noise of the Frio CASSM data allowed for
and stacked in 15 min intervals, for an effective 15 min sampling a lower threshold. Using the numerical reservoir model (Fig. 1) and
rate. The resulting deviations in traveltimes from their pre-injection the seismic velocity structure discussed earlier (Fig. 3), we set up the
values are plotted in Fig. 12. inverse problem. In particular, we define all the quantities needed
552 D.W. Vasco, T.M. Daley and A. Bakulin

to completely specify the data and coefficients in eqs (29) and (30). permeability in the layer, the variations are probably not significant.
Our initial reservoir model consists of a uniform layer 2.5 m thick We should note that the deviations in the ray paths following the
with a porosity of 25 per cent and a horizontal permeability of onset times are much larger and thus should be taken into account
3.0 × 10−12 m2 . The layer dips, as in the synthetic example, but when considering later traveltime changes.
we now account for the associated density effects. The relative per- The initial and final calculated traveltime histories are shown in
meability and capillary pressure behaviour were taken from earlier Fig. 14. The final calculated onset times are in fair agreement with
studies (Doughty et al. 2008; Daley et al. 2011). The approach of the observed values, as indicated in Fig. 15. For example, the fi-
Hill (Hill 1963) was used to relate the saturation changes to velocity nal onset time at the shallowest receiver (1654 m) is slightly less
changes in the poroelastic model. For each iteration of the inver- then 2 d. The arrival time is compatible with the carbon dioxide
sion algorithm, the trajectories and related sensitivities are based breakthrough time obtained from the U-tube sampler (Daley et al.
on the results of a reservoir simulation with updated reservoir prop- 2011). The onset time for the receiver at 1658 m depth has also
erties. The final misfit obtained after three iterations, around 0.1 d, advanced in time to about 0.7 d, in better agreement with the ob-
is similar to that obtained in the synthetic test. The final model of served value. Even though the magnitudes are not matched, there
permeability variation between the wells is plotted in Fig. 13. There is qualitative agreement between the final predicted and the actual

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

are generally higher permeabilities in the region between the wells. observed traveltime changes.
The drop off in permeability near the monitoring well is probably
an artifact of poor sampling at that end. In Fig. 13 we have also
plotted the all the ray paths from the start of injection until the 4 C O N C LU S I O N S
onset time at each receiver. The variation in ray paths in Fig. 13
Relating changes in reservoir properties to changes in geophysi-
allows us to examine any deviations from the assumption that the
cal observations can be a difficult task, often complicated by the
propagation does not change between the start of injection and the
fact that time-lapse changes are typically severely under-sampled in
onset time. In general this appears to be true for the source–receiver
time. Thus, temporal changes are almost always aliased and distinct
pairs used here. However, for two source–receiver pairs we observe
processes, such as fluid pressure changes and saturation changes can
some variation in ray paths. Given the 3 m discretization of the
be confused. Such a confounding of signals can occur even though
these processes operate on distinctly different time scales. For ex-
ample, at the Frio site transient pressure changes can propagate
between the wells in less then a minute while saturation changes
may take more than a day to travel the same distance. This points to
the importance of adequate temporal sampling to better understand
fluid flow in the subsurface. With sufficient sampling in time, it is
possible to estimate the onset of a change in seismic arrival time for
a given source–receiver pair. Such onset times have a rather direct
relationship to flow-related changes in the state of a reservoir. For
example, at the Frio site the initiation of changes in seismic arrival
times is related to fluid saturation breakthrough times. Onset times
are less influenced by the rock physics model relating fluid satura-
tion changes to changes in seismic velocity. For this reason, onset
times provide a robust basis for reservoir characterization based
upon geophysical observations.
As the development of the CASSM system indicates (Ajo-
Franklin et al. 2011), it is certainly possible to improve the temporal
sampling associated with the geophysical monitoring of flow pro-
cesses. Improvements in temporal sampling are also on the horizon
for time-lapse seismic reflection monitoring (Bakulin et al. 2012). It
is hoped that these enhancements can continue in better constrained
settings, such as full crosswell and surface seismic reflection sur-
veys. The approach described in this paper can be a tool for data
reduction, allowing for the extraction of a 3-D field of onset times
from a 4-D time lapse data set. Finally, onset times also provide a
means for combining multiple geophysical data sets, such as seismic
and ground penetrating radar observations. Such combinations of
data can better constrain reservoir geometry (Doetsch et al. 2010b)
and may be combined with hydrological data for improved aquifer
characterization (Lochbuhler et al. 2013).

AC K N OW L E D G E M E N T S

Figure 13. Estimated permeability variations between the injection and Work performed at Lawrence Berkeley National Laboratory was
observation wells based upon the inversion of data from the Frio II pilot supported by Saudi Aramco and by the US Department of En-
experiment. The ray paths for the six receivers and for all times up to the ergy under contract number DE-AC02-05- CH11231, Office of Ba-
onset time are also plotted here. sic Energy Sciences, and the GEOSEQ project for the Assistant
 Utilizing the onset of time-lapse changes 553

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

Figure 14. Traveltime histories calculated using the initial and final reservoir models of the iterative inversion algorithm.

Amelung, F., Galloway, D.L., Bell, J.W., Zebher, H.A. & Laczniak, R.J.,
1999. Sensing the ups and downs of Las Vegas – InSAR reveals structural
control of land subsidence and aquifer-system deformation, Geology, 27,
483–486.
Arts, R., Eiken, O., Chadwick, A., Zweigel, P., van der Meer, L. & Zinszner,
B., 2004. Monitoring of CO2 injected at Sleipner using time-lapse seismic
data, Energy, 29, 1383–1392.
Bakulin, A., Burnstad, R., Jervis, M. & Kelamis, P., 2012. Evaluating
permanent seismic monitoring with shallow buried sensors in a desert
environment, SEG Technical Program Expanded Abstracts, Las Vegas,
1–5.
Behrens, R., Condon, P., Haworth, W., Bergeron, M., Wang, Z. & Ecker, C.,
2002. Monitoring of water influx at Bay Marchand: the practical use of
4D in an imperfect world, SPE Res. Eval. Eng., 5(5), 410–420.
Benson, R.D. & Davis, T.L., 2000. Time-lapse seismic monitoring and dy-
namic reservoir characterization, Central Vacuum Unit, Lea County, New
Mexico, SPE Res. Eval. Eng., 3, 88–97.
Burkhart, T., Hoover, A.R. & Flemings, P.B., 2000. Time-lapse (4D) seismic
monitoring of primary production of turbidite reservoirs as South Tim-
balier Block 295, offshore Louisiana, Gulf of Mexico, Geophysics, 65,
351–367.
Camporese, M., Cassiani, G., Deiana, R. & Salandin, P., 2011. Assess-
ment of local hydraulic properties from electrical resistivity tomography
Figure 15. Plot of the initial fit to the observed onset times (open circles) monitoring of a three-dimensional synthetic tracer test experiment, Water
and the final fit (filled squares) based upon the permeability model plotted Resour. Res., 47, W12508, doi:10.1029/2011WR010528.
in Fig. 13. For an ideal fit the points would lie upon the 45◦ line plotted in Castillo, W., Hunter, S., Harben, P., Wright, C., Conant, R. & Davis, E.,
the figure. 1997. Deep hydraulic fracture imaging: recent advances in tiltmeter tech-
nologies, Int. J. Rock. Mech. Min. Sci., 34, 3–4.
Secretary for Fossil Energy, Office of Coal and Power Systems, Chadwick, R.A., Williams, G.A., Williams, J.D.O. & Noy, D.J., 2012. Mea-
through the National Energy Technology Laboratory of the US De- suring pressure performance of a large saline aquifer during industrial-
partment of Energy. scale CO2 injection: the Utsira Sand, Norwegian North Sea, Int. J. Green-
house Gas Cont., 10, 374–388.
Chapman, C.H., 2004. Fundamentals of Seismic Wave Propagation, Cam-
bridge Univ. Press.
REFERENCES Coscia, I., Greenhalgh, S., Linde, N., Doetsch, J., Marescot, M., Gunther,
Ajo-Franklin, J., Daley, T., Butler-Veytia, B., Peterson, J., Wu, Y., Kelly, T., Vogt, T. & Green, A.G., 2011. 3D crosshole ERT for aquifer char-
B. & Hubbard, S., 2011. Multi-level continuous active source seis- acterization and monitoring of infiltrating river water, Geophysics, 76,
mic monitoring (ML-CASSM): mapping shallow hydrofracture evolu- G49–G59.
tion at a TCE contaminated site, SEG Expanded Abstracts, 30, 3727, Coscia, I., Linde, N., Greenhalgh, S., Vogt, T. & Green, A., 2012. Estimating
doi:10.1190/1.3627980. traveltimes and groundwater flow patterns using 3D time-lapse crosshole
554 D.W. Vasco, T.M. Daley and A. Bakulin

ERT imaging of electrical resistivity fluctuations induced by infiltrating Johnston, D.H., McKenny, R.S., Verbeek, J. & Almond, J., 1998. Time-lapse
river water, Geophysics, 77, E239–E250. seismic analysis of Fulmar Field, Leading Edge, 17, 1420–1428.
Daley, T.M., Solbau, R.D., Ajo-Franklin, J.B. & Benson, S.M., 2007. Con- Karaoulis, M., Revil, A., Werkema, D.D., Minsley, B., Woodruff, W.F. &
tinuous active-source monitoring of CO2 injection in a brine aquifer, Kemna, A., 2011. Time-lapse 3D inversion of complex conductivity data
Geophysics, 72, A57–A61. using an active time constrained (ATC) approach, Geophys. J. Int., 187,
Daley, T.M., Myer, L.R., Peterson, J.E., Majer, E.L. & Hoversten, G.M., 237–251.
2008. Time-lapse crosswell seismic and VSP monitoring of injected CO2 Kowalsky, M.B., Finsterle, S. & Rubin, Y., 2004. Estimating flow parameter
in a brine aquifer, Environ. Geol., 54, 1657–1665. distributions using ground-penetrating radar and hydrological measure-
Daley, T.M., Ajo-Franklin, J.B. & Doughty, C., 2011. Constraining the ments during transient flow in the vadose zone, Adv. Water Resour., 27,
reservoir model of an injected CO2 plume with crosswell CASSM at the 583–599.
Frio-II brine pilot, Int. J. Greenhouse. Gas Cont., 5, 1022–1030. Kowalsky, M.B., Chen, J. & Hubbard, S., 2006. Joint inversion of geo-
Day-Lewis, F., Harris, J.M. & Gorelick, S.M., 2002. Time-lapse inversion physical and hydrological data for improved subsurface characterization,
of cross-well radar data, Geophysics, 67, 1740–1752. Leading Edge, 25, 730–734.
Doetsch, J., Linde, N. & Binley, A., 2010a. Structural joint inversion of LaBrecque, D.J. & Yang, X., 2001. Difference inversion of ERT data: a fast
time-lapse crosshole ERT and GPR traveltime data, Geophys. Res. Lett., inversion method for 3-D in situ monitoring, J. Environ. Eng. Geophys.,
37, L24404, doi:10.1029/2010GL045482. 6, 83–89.

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

Doetsch, J., Linde, N., Coscia, I., Greenhalgh, S.A. & Green, A.G., 2010b, Landa, J.L. & Horne, R.N., 1997. A procedure to integrate well test data,
Zonation for 3D aquifer characterization based on joint inversions of reservoir performance history, and 4-D seismic information into a reser-
multimethod crosshole geophysical data, Geophysics, 75, G53–G64. voir description, in Proceedings of the SPE Annual Technical Conference
Doughty, C., Freifeld, B.M. & Trautz, R.C., 2008. Site characterization for and Exhibition, October 5–8, San Antonio, Texas.
CO2 geologic storage and vice versa: the Frio brine pilot, Texas, USA as Landro, M., 2001. Discrimination between pressure and fluid saturation
a case study, Environ. Geol., 54, 1635–1656. changes from time-lapse seismic data, Geophysics, 66, 836–844.
Du, J. & Olson, J.E., 2001. A poroelastic reservoir model for predicting Lazaratos, S.K. & Marion, B.P., 1997. Crosswell seismic imaging of reser-
subsidence and mapping subsurface pressure fronts, J. Petrol. Sci. Eng., voir changes caused by CO2 injection, Leading Edge, 16, 1300–1307.
30, 181–197. Lee, D.S., Stevenson, V.M., Johnston, P.F. & Mullen, C.E., 1995, Time-
Eastwood, J., Lebel, J.P., Dilay, A. & Blakeslee, S., 1994. Seismic monitoring lapse crosswell seismic tomography to characterize flow structure in the
of steam-based recovery of bitumen, Leading Edge, 4, 242–251. reservoir during thermal stimulation, Geophysics, 60, 660–666.
Evans, K.F., Holzhausen, G.R. & Wood, M.D., 1982. The geometry of a Lochbuhler, T., Doetsch, J., Brauchler, R. & Linde, N., 2013. Structure-
large-scale nitrogen gas hydraulic fracture formed in Devonian shale: an coupled joint inversion of geophysical and hydrological data, Geophysics,
example of fracture mapping using tiltmeters, Soc. Petrol. Eng. J., 22, 78, ID1–ID14.
755–763. MacBeth, C. & Al-Maskeri, Y., 2006. Extraction of permeability from time-
Fielding, E.J., Blom, R.G. & Goldstein, R.M., 1998. Rapid subsidence over lapse seismic data, Geophys. Prospect., 54, 333–349.
oil fields measured by SAR interferometry, Geophys. Res. Lett., 25, 3215– Martinez-Pagan, P., Jardani, A., Revil, A. & Haas, A., 2010. Self-potential
3218. monitoring of a salt plume, Geophys. J. Int., 75, WA17–WA25.
Freifeld, B.M., Trautz, R.C., Kharaka, Y.K., Phelps, T.J., Myer, L.R., Hov- Mathisen, M.E., Vasiliou, A.A., Cunningham, P., Shaw, J., Justice, J.H.
orka, S.D. & Collins, D.J., 2005. The U-tube: a novel system for acquiring & Guinzy, N.J., 1995. Time-lapse crosswell seismic tomogram inter-
borehole fluid samples from a deep geologic CO2 sequestration experi- pretation: implications for heavy oil reservoir characterization, thermal
ment, J. geophys. Res., 110, B10203, doi:10.1029/2005JB003735. recovery process monitoring, and tomographic imaging technology, Geo-
Gassmann, F., 1951. Ueber die elastizitat poroser medien, Vier. Der. Natur., physics, 60, 631–650.
Heft, 96, 1–23. Mavko, G., Mukerji, T. & Dvorkin, J., 1998. The Rock Physics Handbook,
Ghosh, R. & Sen, M.K., 2012. Predicting subsurface CO2 movement: from Cambridge Univ. Press.
laboratory to field scale, Geophysics, 77, M27–M37. McKenna, J., Sherlock, D. & Evans, B., 2001. Time-lapse 3-D seismic
Gosselin, O. et al., 2003. History matching using time-lapse seismic imaging of shallow subsurface contaminant flow, J. Contam. Hydrol., 53,
(HUTS), SPE Paper, 84464, doi:10.2118/84464-MS. 133–150.
Greaves, R.J. & Fulp, T., 1987. Three dimensional seismic monitoring of an Menke, W., 1989. Geophysical Data Analysis: Discrete Inverse Theory,
enhanced oil recovery process, Geophysics, 52, 1175–1187. Academic Press.
Guilbot, J. & Brackin, S., 2002. 4-D constrained depth conversion for reser- Miller, C.R., Routh, P.S., Brosten, T.R. & McNamara, J.P., 2008. Application
voir compaction estimation: application to Ekofisk field, Leading Edge, of time-lapse ERT to watershed characterization, Geophysics, 73, G7-
21, 302–308. G17.
Hashin, Z. & Shtrikman, S., 1963. A variation approach to the elastic be- Mossop, A. & Segall, P., 1999. Volume strain within the Geysers geothermal
havior of multiphase materials, J. Mech. Phys. Solids, 11, 127–140. field, J. geophys. Res., 104, 29113–29131.
He, W., Guerin, G., Anderson, R.N. & Mello, U.T., 1998. Time-dependent Müller, T.M., Gurevich, B. & Lebedev, M., 2010. Seismic wave attenua-
reservoir characterization of the LF sand in the South Eugene Island 330 tion and dispersion resulting from wave-induced flow in porous rocks—a
Field, Gulf of Mexico, Leading Edge, 17, 1434–1438. review, Geophysics, 75, A147–A164.
Hill, R., 1963. Elastic properties of reinforced solids: some theoretical prin- Noble, B. & Daniel, J.W., 1977. Applied Linear Algebra, Prentice-Hall.
ciples, J. Mech. Phys. Solids, 11, 357–372. Oppliger, G., Coolbaugh, M. & Shevenell, L., 2006. Improved visualization
Hodgson, N., MacBeth, C., Duranti, L., Rickett, J. & Nihei, K., 2007. Invert- of satellite radar InSAR observed structural controls at producing geother-
ing for reservoir pressure change using time-lapse time strain: application mal fields using modeled horizontal surface displacements, Geotherm.
to the Genesis field, Gulf of Mexico, Leading Edge, 26, 649–652. Res. Coun. Trans., 30, 927–930.
Hoversten, G.M., Gritto, R., Washbourne, J. & Daley, T., 2003. Pressure and Osdal, B., Husby, O., Aronsen, H.A., Chan, N. & Alsos, T., 2006. Mapping
fluid saturation prediction in a multicomponent reservoir using combined the fluid front and pressure buildup using 4D data on Norne Field, Leading
seismic and electromagnetic imaging, Geophysics, 68, 1580–1591. Edge, 25, 1134–1141.
Huang, X., Meister, L. & Workman, R., 1998. Improving production history Paige, C.C. & Saunders, M.A., 1982. LSQR: An algorithm for sparse lin-
matching using time-lapse seismic data, Leading Edge, 17, 1430–1433. ear equations and sparse least squares, ACM Trans. Math. Software, 8,
Jandani, A. et al., 2009. Reconstruction of the water table from self-potential 43–71.
data: a Bayesian approach, Groundwater, 47, 213–227. Palmer, I.D., 1990. Uplifts and tilts at Earth’s surface induced by pressure
Johnson, D.L., 2001. Theory of frequency dependent acoustics in patchy transients from hydraulic fractures, Soc. Petrol. Eng. Product. Eng., 5,
saturated porous media, J. acoust. Soc. Am., 110, 682–694. 324–332.
 Utilizing the onset of time-lapse changes 555

Peaceman, D.W., 1977. Fundamentals of Numerical Reservoir Simulation, Vasco, D.W., Yoon, S. & Datta-Gupta, A., 1999. Integrating dynamic data
Elsevier Scientific Publishing. into high-resolution reservoir models using streamline-based analytic sen-
Podvin, P. & Lecomte, I., 1991. Finite difference computation of traveltimes sitivity coefficients, Soc. Petrol. Eng. J., 4(4), 389–399.
in very contrasted velocity models: a massively parallel approach and its Vasco, D.W., Karasaki, K. & Doughty, C., 2000. Using surface deformation
associated tools, Geophys. J. Int., 105, 271–284. to image reservoir dynamics, Geophysics, 65, 132–147.
Pride, S.R., Berryman, J.G. & Harris, J.M., 2004. Seismic attenu- Vasco, D.W., Karasaki, K. & Kishida, K., 2001. A coupled inversion of
ation due to wave-induced flow, J. geophys. Res., 109, B01201, pressure and surface displacement, Water Resour. Res., 37, 3071–3089.
doi:10.1029/2003JB002639. Vasco, D.W., Karasaki, K. & Nakagome, O., 2002. Monitoring production
Pruess, K., 2004. The TOUGH Codes—a family of simulation tools for using surface deformation: the Hijiori test site and the Okuaizu geothermal
multiphase flow and transport processes in permeable media, Vadose field, Japan, Geothermics, 31, 303–342.
Zone J., 3, 738–746. Vasco, D.W., Datta-Gupta, A., Behrens, R., Condon, P. & Rickett, J.,
Pruess, K., Oldenburg, C. & Moridis, G., 1999. TOUGH2 User’s Guide, 2004. Seismic imaging of reservoir flow properties: Time-lapse ampli-
Version 2.0, LBNL Report, 43134, Berkeley, 1999. tude changes, Geophysics, 69(6), 1425–1442.
Rucci, A., Vasco, D.W. & Novali, F., 2010. Fluid pressure arrival time Vasco, D.W., Rucci, A., Ferretti, A., Novali, F., Bissell, R., Ringrose, P.,
tomography: estimation and assessment in the presence of inequality Mathieson, A. & Wright, I., 2010. Satellite-based measurements of sur-
constraints, with an application to production at the Krechba field, Algeria, face deformation reveal fluid flow associated with the geological storage

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020

Geophysics, 75, O39–O55. of carbon dioxide, Geophys. Res. Lett., 37, doi:10.1029/2009GL041544,
Rucci, A., Vasco, D.W. & Novali, F., 2013. Monitoring the geological storage 1–5.
of carbon dioxide using multicomponent SAR interferometry, Geophys. Vidale, J., 1988. Finite-difference calculation of travel times, Bull. seism.
J. Int., 193, 197–208. Soc. Am., 78, 2062–2076.
Schmidt, D.A. & Burgmann, R., 2003. Time-dependent land uplift and Wang, Z., Cates, M.E. & Langan, R.T., 1998. Seismic monitoring of a
subsidence in the Santa Clara valley, California, from a large inter- CO2 flood in carbonate reservoir: a rock physics study, Geophysics, 63,
ferometric synthetic aperture radar data set, J. geophys. Res., 108, 1604–1617.
doi:10.1029/20002JB002267. Watts, G.F.T., Jizba, D., Gawith, D.E. & Gutteridge, P., 1996. Reservoir
Shapiro, S.A., Audigane, P. & Royer, J.-J., 1999. Large-scale in situ per- monitoring of the Magnus field through 4D time-lapse seismic analysis,
meability tensor from induced microseismicity, Geophys. J. Int., 137, Petrol. Geosci., 2, 361–372.
207–213. White, J.E., 1975. Computed seismic speeds and attenuation in rocks with
Shapiro, S.A., Rothert, E., Rath, V. & Rindschwentner, J., 2002. Characteri- partial gas saturation, Geophysics, 40, 224–232.
zation of fluid transport properties of reservoirs using induced microseis- Wiener, O., 1910. Zur theorie der refraktionskonstanten, Berichteuber Ver-
micity, Geophysics, 67, 212–220. handlungen Koniglich-Sachsischen Gesellschaft Wisseschaften Leipzig,
Shi, J.-Q., Xue, Z. & Durucan, S., 2007. Seismic monitoring and modelling 62, 256–277.
of supercritical CO2 injection into a water-saturated sandstone: interpre- Wright, C.A., 1998. Tiltmeter fracture mapping: from the surface and now
tation of P-wave velocity data, Int. J. Greenhouse Gas Cont., 1, 473– downhole, Petrol. Eng. Int., January, 50–63.
480. Wright, C.A., Davis, E.J., Minner, W.A., Ward, J.F., Weijers, L., Schell,
Smith, P., Berg, J.I., Eidsvig, S., Magnus, I., Verhelst, F. & Helgesen, J., E.J. & Hunter, S.P., 1998. Surface tiltmeter fracture mapping reaches
2001. 4-D seismic in a complex fluvial reservoir: The Snorre feasibility new depths - 10,000 feet and beyond?, Soc. Petrol. Eng., April, 39919,
study, Leading Edge, 20, 270–276. doi:10.2118/39919-MS.
Stancliffe, R.P.W. & van der Kooij, M.W.A., 2001. The use of satellite-based Zhang, F., Juhlin, C., Cosma, C., Tryggvason, A. & Pratt, R.G., 2012. Cross-
radar interferometry to monitor production activity at the Cold Lake well seismic waveform tomography for monitoring CO2 injection: a case
heavy oil field, Alberta, Canada, Am. Assoc. Petrol. Geol. Bull., 85, 781– study from the Ketzin Site, Germany, Geophys. J. Int., 189, 629–646.
793. Zhang, J., Revil, A., Karaoulis, M. & Vasco, D., 2014. Crosswell 4D resistiv-
Teatini, P. et al., 2011. Geomechanical response to seasonal gas storage in ity tomography localized the oil water encrouchment front during water
depleted reservoirs: A case study in the Po River basin, Italy, J. geophys. flooding, Geophysics, in press.
Res., 116, 1–21.
Tong, F., Jing, L. & Zimmerman, R.W., 2009. An effective thermal con-
ductivity model of geological porous material for coupled thermo-hydro- APPENDIX A
mechanical systems with multiphase flow, Int. J. Rock Mech. Mining Sci.,
46, 1358–1369.
In this Appendix, we follow Peaceman (1977, p. 19) and derive the
Tura, A. & Lumley, D.E., 1998. Subsurface fluid-flow properties from time- relationship between the velocity of the non-aqueous phase (vn ) and
lapse elastic-wave reflection data, in Proceedings of 43rd SPIE Annual the total velocity (vt ), which is defined as
Meeting, July 19, San Diego, CA, 125–138.
vt = vw + vn . (A1)
Vasco, D.W., 2004a, Seismic imaging of reservoir properties: Time-lapse
pressure changes, Geophysics, 69, 511–521. From the definition of capillary pressure, eq. (2), we have
Vasco, D.W., 2004b. Estimation of flow properties using surface deforma-
tion and head data: A trajectory-based approach, Water Resour. Res., 40, ∇ pc = ∇ pn − ∇ pw . (A2)
W10104, doi:10.1029/2004WR003272, 1–14. If we define the phase mobilities
Vasco, D.W., 2011. On the propagation of a coupled saturation and pressure
front, Water Resour. Res., 47, 1–21. K kr w
λw = (A3)
Vasco, D.W. & Datta-Gupta, A., 1999. Asymptotic solutions for solute trans- μw
port: A formalism for tracer tomography, Water Resour. Res., 35, 1–16.
and
Vasco, D.W. & Datta-Gupta, A., 2001. Asymptotics, saturation fronts, and
high resolution reservoir characterization, Transport Porous Med., 42, K kr n
λn = , (A4)
315–350. μn
Vasco, D.W. & Ferretti, A., 2005. On the use of quasi-static deformation to we can write the fluid phase velocities as
understand reservoir fluid flow, Geophysics, 70, O13–O27.
Vasco, D.W. & Finsterle, S., 2004. Numerical trajectory calculations for the vw = −λw (∇ pw − ρw gZ) (A5)
efficient inversion of transient flow and tracer observations, Water Resour.
Res., 40, 1–17. vn = −λn (∇ pn − ρn gZ) . (A6)
556 D.W. Vasco, T.M. Daley and A. Bakulin

Rearranging and combining eqs (A2), (A5) and (A6) gives vn = f n vt − λw f n [∇ pc + (ρw − ρn ) gZ] . (A9)
(λn + λw ) vn = λn vt − λn λw [∇ pc + (ρw − ρn ) gZ] (A7) Treating the capillary pressure as a function of Sn we can write (A9)
as
if we use eq. (A1) to eliminate vw . Dividing through by λn + λw and  
defining the fractional flow function for the non-aqueous phase d pc
vn = f n vt − λw f n ∇ Sn + (ρw − ρn ) gZ , (A10)
d Sn
λn
fn = (A8) and so relate the velocity of the non-aqueous phase to the fractional
λn + λw
flow function fn , the total fluid velocity, the capillary pressure, the
we can write (A7) as
saturation Sn , and the density difference of the two fluids.

Downloaded from https://academic.oup.com/gji/article-abstract/197/1/542/687746 by guest on 21 May 2020