4th International FLAC Symposium on Numerical Modeling in Geomechanics 2006 Hart & Varona (eds.

) Paper: 04-10
2006 Itasca Consulting Group, Inc., Minneapolis, ISBN 0-9767577-0-2

Earthquake deconvolution for FLAC

L.H. Mejia

URS Corporation, Oakland, CA, USA

E.M. Dawson

URS Corporation, Los Angeles, CA, USA

ABSTRACT: Design earthquake ground motions for dynamic analyses are typically specified as outcrop motions, which may have to be modified for input at the base of a FLAC model. Often a deconvolution analysis
using a 1-D wave propagation code, such as the program SHAKE, is performed to obtain the appropriate input motion at depth. This seemingly simple analysis is often the subject of considerable confusion. In this paper the theory and operation of the program SHAKE and input requirements of FLAC are reviewed, and the
application of SHAKE for adapting design earthquake motions for FLAC input is described. Numerical examples illustrating typical cases are presented, and several questions that commonly arise are addressed.
1 INTRODUCTION

grid points along the base of the mesh. While simple

to use, a potential drawback of a rigid base is that
the motion at the base of the model is completely
prescribed. Hence, the base acts as if it were a fixeddisplacement boundary reflecting downward propagating waves back into the model. Thus, a rigid base
is not an appropriate boundary for general application unless a large dynamic impedance contrast is
meant to be simulated at the base (e.g. low velocity
sediments over high velocity bedrock).

Design earthquake ground motions developed for

seismic analyses are usually provided as outcrop
motions often rock outcrop motions. However, for
FLAC analyses, seismic input must be applied at the
base of the model rather than at the ground surface
as illustrated in Figure 1. The question then arises:
what input motion should be applied at the base of a
FLAC model in order to properly simulate the design
motion?
The appropriate input motion at depth can be
computed through a deconvolution analysis using a
1-D wave propagation code such as the equivalentlinear program SHAKE. This seemingly simple
analysis is often the subject of considerable confusion resulting in improper ground motion input for
FLAC models. In this paper the application of
SHAKE for adapting design earthquake motions for
FLAC input is described. Numerical examples are
presented illustrating two typical cases:
1 A rigid base, where an acceleration-time history
is specified at the base of the FLAC mesh.
2 A compliant base, where a quiet (absorbing)
boundary is used at the base of the FLAC mesh.

Target earthquake given

as outcrop motion

Soil
Bedrock

Input
Motion?

Figure 1. Seismic input to FLAC.

For a compliant base simulation, a quiet (also referred to as absorbing) boundary is specified along
the base of the FLAC mesh. FLAC uses the viscous
boundary scheme developed by Lysmer & Kuhlemeyer (1969), consisting of two sets of dashpots attached independently to the mesh in the normal and
shear directions. Limiting discussion to the shear di-

2 SEISMIC INPUT TO FLAC

Input of an earthquake motion into FLAC is typically done using either a rigid base or a compliant
base. For a rigid base, a time-history of acceleration (or velocity or displacement) is specified for
1

rection only, the dashpots provide a viscous shear

traction given by:

s = Cs vs

interface between two elastic materials (Kolsky

1963) can be expressed in terms of recursion formulas. In this way, the upward and downward propagating motions in one layer can be computed from
the upward and downward motions in a neighboring
layer.
To satisfy the zero shear stress condition at the
free surface, the upward and downward propagating
motions in the top layer must be equal. Starting at
the top layer, repeated use of the recursion formulas
allows the determination of a transfer function between the motions in any two layers of the system.
Thus, if the motion is specified at one layer in the
system, the motion at any other layer can be computed.
SHAKE input and output is not in terms of the
upward and downward propagating wave trains, but
in terms of the motions at: a) the boundary between
two layers, referred to as a within motion, or b) at
a free surface, referred to as an outcrop motion.
The within motion is the superposition of the upward and downward propagating wave trains. The
outcrop motion is the motion that would occur at a
free surface at that location. Hence the outcrop motion is simply twice the upward propagating wave
train motion. If needed, the upward propagating motion can be computed by taking half the outcrop motion. At any point, the downward propagating motion can then be computed by subtracting the upward
propagating motion from the within motion.
This SHAKE solution is in the frequency domain,
with conversion to and from the time-domain performed with a Fourier transform. Although this paper is only concerned with the use of SHAKE for the
linear elastic case, SHAKE can address non-linear
soil behavior approximately through the equivalentlinear approach. Analyses are run iteratively to obtain shear modulus and damping values for each
layer that are compatible with the computed effective strain for the layer.

(1)

where and CS are the density and shear wave velocity of the base material, and vS is the shearcomponent of particle velocity at the boundary. Note
that equation (1) is simply the relation between shear
stress and particle velocity in an elastic shear wave
(Kolsky 1963). The viscous dashpots of the quiet
boundary absorb downward propagating waves so
that they are not reflected back into the model.
At a quiet boundary, an acceleration time history
cannot be input directly because the boundary must
be able to move freely to absorb incoming waves.
Instead the acceleration-time history is transformed
into a stress-time history for input. First the acceleration is integrated to obtain velocity and then the
proportionality of stress to velocity in an elastic
wave is used, as in Equation (1).
FLAC input requires that a factor of two be added
to this relation because of the stress is absorbed
by the viscous dashpots. FLAC does not take care of
this numerical detail internally, but instead requires
the user to add the factor of two. Thus for application of a stress-time history through a quiet base, the
shear stress is given by

s = 2 C s v su

(2)

where vsu is the particle velocity of the upward

propagating motion. Note that if a history of acceleration is recorded at a grid point on the quiet base,
it will not necessarily match the input history. The
input stress-time history specifies the upward propagating wave motion into the FLAC model, but the
actual motion at the base will be the superposition of
the upward motion and the downward motion reflected back from the FLAC model.

3 THEORY AND OPERATION OF SHAKE

Free surface

SHAKE (Schnabel et al. 1972) is a widely used 1-D

wave propagation code for site response analysis.
SHAKE computes the vertical propagation of shear
waves through a profile of horizontal visco-elastic
layers. Within each layer, the solution to the wave
equation can be expressed as the sum of an upward
propagating wave train and a downward propagating
wave train. The SHAKE solution is formulated in
terms of these upward and downward propagating
motions within each layer as illustrated in Figure 2.
The relation between waves in one layer and
waves in an adjacent layer can be solved by enforcing the continuity of stresses and displacements at
the interface between the layers. These well known
relations for reflected and transmitted waves at the

Layer 1

G1 1 1

Layer 2

G2 2 2

Layer n
(halfspace)

Gn n n
Downward propagating
Upward propagating

Figure 2. Layered system analyzed by SHAKE. Layer properties are shear modulus, G; density ; and damping fraction, .
2

The SHAKE model includes the three elastic layers and an elastic halfspace with the same properties
as the bottom layer. The FLAC model consists of a
column of 120 linear elastic elements. The target
earthquake is input at the top of the SHAKE column
as an outcrop motion. Then, the motion at the top of
the halfspace is extracted as a within motion
(shown in Fig. 5) and is applied as an accelerationtime history to the base of the FLAC model. The resulting acceleration at the surface of the FLAC
model is shown to be virtually identical to the target
motion in Figure 6. The SHAKE within motion is
appropriate for rigid base input because, as described above, the within motion is the actual motion at that location, the superposition of the upward
and downward propagating waves.

4 DECONVOLUTION FOR RIGID AND

COMPLIANT BASE
The deconvolution procedure for a rigid base is illustrated in Figure 3. The goal of the exercise is to
determine the appropriate base input motion to
FLAC such that the target design motion is recovered at the top surface of the FLAC model. The profile modeled consists of three 20-m thick elastic layers with shear wave velocities and densities as
shown in the figure. A nominal material damping of
0.1% is used for all layers in order to minimize approximations introduced by the Rayleigh damping
model employed in the FLAC analysis (Rayleigh
damping is frequency dependent). The target acceleration-time history, shown in Figure 4, is a modified Kobe Earthquake recording, scaled to a peak
ground acceleration (PGA) of 1.0 g.

SHAKE

FLAC
Target earthquake
applied at surface as
outcrop motion

+0 m
vs = 150 m/sec
= 18 kN/m3

Computed acceleration
record at surface

-20 m
vs = 225 m/sec
= 19 kN/m3
-40 m
vs = 350 m/sec
= 22 kN/m3
-60 m
Halfspace
vs = 350 m/sec
= 22 kN/m3

SHAKE within motion applied

as acceleration time history to
base of FLAC model

Figure 3. Deconvolution procedure for rigid base.

acceleration (g)

1.0
0.5
0.0
Modified Kobe Earthquake
Scaled to PGA = 1.0 g

-0.5
-1.0
0

10

time (sec)

Figure 4. Target earthquake acceleration-time history.

15

20

25

acceleration (g)

0.5
Within Motion
0.0

-0.5
0

10

Figure 5. Computed within motion from SHAKE.

15

time (sec)

20

25

acceleration (g)

1.0
Rigid Base

0.5

Target Motion
FLAC

0.0
-0.5
-1.0
0

10

Figure 6. Computed and target motions: rigid base.

time (sec.)

The deconvolution procedure for a compliant

base is illustrated in Figure 7. The SHAKE and
FLAC models are identical to those for the rigid
body exercise, except that a quiet boundary is applied to the base of the FLAC mesh. For application
through a quiet base, the upward propagating wave
motion ( the outcrop motion) is extracted from
SHAKE at the top of the halfspace. This acceleration-time history (shown in Fig. 8) is integrated to
obtain a velocity, which is then converted to a stress
history using Equation (2). Again, the resulting acceleration at the surface of the FLAC model is virtually identical to the target motion (Fig. 9).
As an additional check of the computed accelerations, the response spectra for both the compliant
base and rigid base cases are shown in Figure 10.
These closely match the response spectra of the target motion.

15

20

25

bedrock. To compute the correct FLAC compliant

base input, a SHAKE model is constructed as shown
in the Figure. The SHAKE model includes a bedrock layer equal in thickness to the elastic base of
the FLAC mesh, and an underlying elastic half-space
with bedrock properties. The target motion is input
to the SHAKE model as an outcrop motion at the top
of the bedrock (point A). Designating this motion as
outcrop means that the upward propagating wave
motion in the layer directly below point A will be set
equal to the target motion. The upward propagating motion for input to FLAC is extracted at Point B
as the outcrop motion.
For the compliant base case there is actually no
need to include the soil layers in the SHAKE model
as these will have no effect on the upward propagating wave train between points A and B. In fact, for
this simple case, it is not really necessary to perform
a formal deconvolution analysis, as the upward
propagating motion at point B will be almost identical to that at point A. Apart from an offset in time,
the only differences will be due to material damping
between the two points, which will generally be
small for bedrock. Thus, for this very common situation, the correct input motion for FLAC is simply
of the target motion. (Note that the upward propagating wave motion must be converted to a stress
time history using equation (2) which includes a factor of 2 to account for the stress absorbed by the viscous dashpots. Alternatively, the target motion can
be directly used with Equation (1).)

5 DECONVOLUTION FOR TYPICAL CASES

ENCOUNTERED IN PRACTICE
Although useful for illustrating the basic ideas behind deconvolution, the example presented above in
Section 4 is not the typical case encountered in practice. More common is the situation shown in Figure
11, where one or more soil layers (expected to behave non-linearly) overly bedrock (assumed to behave linearly). A FLAC model for this case will usually include the soil layers and an elastic base of
4

SHAKE

FLAC
Target earthquake
applied at surface as
outcrop motion

+0 m
vs = 150 m/sec
= 18 kN/m3

Computed acceleration
record at surface

-20 m
vs = 225 m/sec
= 19 kN/m3
-40 m
vs = 350 m/sec
= 22 kN/m3
-60 m

Downward propagating
motion absorbed by
quiet base

Halfspace
vs = 350 m/sec
= 22 kN/m3

SHAKE upward propagating

motion applied to base of
FLAC Model

Figure 7. Deconvolution procedure for compliant base.

acceleration (g)

0.5
Upward Propagating
0.0

-0.5
0

10

time (sec)

15

20

25

Figure 8. Upward propagating motion from SHAKE.

acceleration (g)

1.0
Compliant Base

0.5

Target Motion
FLAC

0.0
-0.5
-1.0
0

10

time (sec)

Figure 9. Computed and target motions: compliant base.

5

15

20

25

2.5
5% damped

pseudo-spectral accel. (g)

2.0

1.5

1.0
Target Motion
FLAC: Compliant Base
FLAC: Rigid Base

0.5

0.0
0.01

0.1

period (sec)

10

Figure 10. Computed and target response spectra.

SHAKE

FLAC

Non-linear soil

Non-linear soil
Target earthquake
applied at top of bedrock
as outcrop motion

A
Bedrock

B
Bedrock

Downward propagating
motion absorbed by
quiet base

SHAKE upward propagating

motion applied to base of
FLAC Model

Figure 11. Compliant base deconvolution procedure for a typical case encountered in practice.

essentially linear. Again the design motion is input

at the top of the bedrock (point A) as an outcrop motion, and the upward propagating motion for input to
FLAC is extracted at point B. As in the previous example, for a compliant base analysis there is no need
to include the soil layers above point B in the
SHAKE model. These layers have no effect on the
upward propagating motion between points A and B.
Unlike the previous case, the upward propagating
motion can be quite different at points A and B, depending on the impedance contrast between the bedrock and linear soil layer. Thus, it is not appropriate
to skip the deconvolution analysis and use the target
motion directly.

For a rigid base analysis, the within motion at

point B is required. Since this within motion incorporates downward propagating waves reflected off
the ground surface, the non-linear soil layers must
be included in the SHAKE model. However, soil
non-linearity will be modeled quite differently in
FLAC and SHAKE. Thus, it is difficult to compute
the appropriate FLAC input motion for a rigid base
analysis.
Another typical case encountered in practice is illustrated in Figure 12. Here, the soil profile is deep,
and rather than extending the FLAC mesh all the
way down to bedrock, the base of the model ends
within the soil profile. Note that the mesh must be
extended to a depth below which the soil response is
6

SHAKE

FLAC

Non-linear soil

Non-linear soil

Linear soil
Linear soil

SHAKE upward propagating

motion applied to base of
FLAC Model

Downward propagating
motion absorbed by
quiet base

Bedrock

Target earthquake
applied at top of bedrock
as outcrop motion

Figure 12. Compliant base deconvolution procedure for another common case encountered in practice.

6 REFLECTIONS OFF RIGID BASE

modes shown in the figure. Note that the boundary

conditions for these waves are fixed displacement at
the base of the model and zero shear stress at the
ground surface.
Also shown in Figures 13 & 14 are the surface
acceleration and response spectra for the corresponding compliant base analysis (velocity in the
FLAC model reduced by 5%). As expected, these
differ only slightly from the target motion with no
signs of standing waves.

One of the main disadvantages of an assumed rigid

base boundary is that downward propagating waves
are reflected back into the model rather than radiating out through the base. These reflections are often
not readily apparent in complex non-linear FLAC
analyses, as they can be masked by the high damping at larger strains in non-linear soil models. Reflections off the base are clearly observed in elastic
systems with very low damping.

6.2 Embankment

6.1 1-D Column

As a further illustration of the spurious reflections

that can be caused by a rigid base, consider the embankment shown in Figure 15. Acceleration input is
applied through a rigid base, using a within motion
computed so that the target earthquake motion is recovered at point A, the free-field. The computed acceleration at the crest of the embankment (Fig. 16)
and the response spectrum of this motion (Fig. 17)
again show large amplitude periodic vibrations. The
corresponding compliant base analysis does not
show these vibrations.

To illustrate how easily unwanted reflections can be

induced off a rigid base, the example shown in Figure 3 is modified slightly. For simplicity all layers
are assigned a uniform shear velocity of 250 m/sec
and uniform density. A SHAKE analysis is then performed to compute the appropriate within motion for
application at the base of the FLAC mesh. Now
imagine that this within motion is applied to a FLAC
model that has a shear wave velocity 5% lower than
the 250 m/sec used in the SHAKE analysis. This
situation might occur, for example, in a 2-D FLAC
analysis where the rock stiffness is slightly lower at
one end of the mesh.
Applying this mismatched rigid-base input motion results in the surface acceleration-time history
shown in Figure 13. Clearly large amplitude periodic
vibrations develop for the rigid-base boundary.
These are due to the excitation of standing waves
within the model. The period of these standing
waves can be seen in the response spectrum (Fig.
14) which has prominent peaks at 0.20 seconds and
0.34 seconds, corresponding to the standing waves

7 CONCLUSIONS
Input of an earthquake motion into FLAC is typically done through either a rigid or compliant base.
For a rigid base, a time-history of acceleration is
specified at the base of the FLAC mesh. For a compliant base, a quiet (or absorbing) boundary is specified along the base of the FLAC mesh and the input
motion is applied as a stress-time history.
7

acceleration (g)

1.5
Rigid Base

1.0

Compliant Base

0.5
0.0
-0.5
-1.0
-1.5
0

10

15

20

25

time (s)

Figure 13. Computed acceleration at top of column for rigid base with 5% velocity mismatch.

pseudo-spectral accel. (g)

Target Motion
Rigid Base
Compliant Base

5% damped

5/4 = H

3/4 = H

4
3
2
1
0
0.01

0.1

10

period (sec)

Figure 14. Response spectrum of surface motion for model with rigid base and 5% velocity mismatch.

Acceleration measured at
crest of embankment
Input motion deconvolved
so as to recover target
motion at free field

1.5
1

100 m

vs = 350 m/sec

100 m

vs = 350 m/sec

Dynamic input applied at rigid base or compliant base

Figure 15. Embankment analyzed with rigid and compliant base.

acceleration (g)

3
Rigid Base

Compliant Base

1
0
-1
-2
-3
0

10

15
time (s)

Figure 16. Computed accelerations at crest of embankment.

20

25

12
Target Motion
Rigid Base
Compliant Base

pseudo-spectral accel. (g)

10
8
6
4
2
0
0.01

0.1

10

period (sec)

Figure 17. Response spectra of motion at crest of embankment.

If the program SHAKE is used to compute the

input motion for application at the base of a FLAC
model, the within motion should be used for a rigid
base, as this is the actual particle motion, the superposition of the upward and downward propagating
wave trains. For a compliant base, the upward
propagating wave train should be used. The upward
propagating wave is the SHAKE outcrop motion.
A rigid base is only appropriate for cases with a
large impedance contrast at the base of the model. A
compliant base is almost always the preferred option
because downward propagating waves are absorbed,
while for a rigid base these waves are reflected back
into the model. Although the presence of these reflections is not always obvious in complex nonlinear FLAC analyses, the can have a major impact
on analysis results, especially when cyclic degradation or liquefaction soil models are employed.
In addition to preventing reflections off the base
of the model, a compliant base greatly simplifies
computation of the appropriate input motion. Only
the upward propagating motion is required. This incoming or incident wave is not affected by the material above. In contrast, a rigid base requires that the
within motion be applied, which is a motion that depends not only on the incoming wave train, but also
on the dynamic response of the model above.

Schnabel, P.B, Lysmer, J. and Seed, B.H. 1972. SHAKE, a

computer program for earthquake response analysis of
horizontally layered sites. Earthquake Engineering Research Center, University of California, Berkeley. Report
No. EERC 72-12.

REFERENCES
Itasca Consulting Group. 2005. FLAC Fast Lagrangian
Analysis of Continua, Ver. 5.0 Users Guide. Minneapolis:
Itasca.
Kolsky, H. 1963. Stress Waves in Solids. New York: Dover
Publications.
Lysmer, J. and Kuhlemeyer, R.L. 1969. Finite Difference
Model for Infinite Media. J. Eng. Mech., 95 (EMR), pp.
859-877.
9