HIGH-PERFORMANCE

COMPUTING

TRACKED VEHICLE SIMULATIONS

AND SEISMIC WAVEFIELD SYNTHESIS
IN SEISMIC SENSOR SYSTEMS
A finite-difference time-domain (FDTD) method models moving vehicle ground motion in
3D geologic environments, producing results similar to those from actual field experiments.
Simulations provide a low-cost alternative to traditional prototype development schemes,
which rely on expensive field tests.

S
ituational awareness is critical to tional awareness on the battlefield. (See the
achieving and maintaining battlefield “Seismic-Wave Propagation” sidebar for fac-
superiority. Given the battlefield’s tors affecting wave propagation and their use
complex and dynamic environmental in vehicle movement research.)
conditions, such awareness requires a spectrum We’ve developed a variable grid finite-difference
of sensor assets to provide necessary informa- time-domain (FDTD) method to model source-spe-
tion with appropriate reliability levels. Seismic cific ground motion attributed to moving vehicles in
sensors implemented in unattended ground realistic 3D geologic environments. Our FDTD-
sensors (UGS) can exploit seismic waves gen- based approach, together with the ready availability
erated from moving wheeled or tracked targets of massively parallel computational support, lets us
and provide detection, location, and discrimi- include all the key target and environmental effects
nation information for targets of interest. 1 observed in real-world seismic signatures. Accurate
Seismic waves carry source-dependent spectral synthesis of seismic data from synthetic tracked and
energy that allows for target discrimination wheeled vehicles lets us supplement real field data
and identification. As Figure 1 illustrates, this inexpensively, predict sensor-system performance,
spectral energy appears as harmonic lines in and optimize system performance, and can serve as
spectrograms, which represent the evolution of ground truth in battlefield-decision-aid technolo-
the signal power’s spectral components over gies concerned with sensor-network deployment.
time. Target signature harmonics depend on From a hardware development perspective, next-
vehicle suspension characteristics, speed, and generation sensor networks will be much more
surface roughness. In a military context, this complex than traditional standalone sensors. Fur-
provides key information for real-time situa- thermore, because of the high sensitivity of seismic
and acoustic signals to environmental variations,
these sensors will require extensive evaluation in a
1521-9615/04/$20.00 © 2004 IEEE
Copublished by the IEEE CS and the AIP much wider variety of environmental conditions
than previous-generation sensors. Hence, a proto-
THOMAS S. ANDERSON, MARK L. MORAN, type development process relying exclusively on
the traditional iterative loop of field-experiment-
STEPHEN A. KETCHAM, AND JAMES LACOMBE
and-laboratory-hardware-modification will likely
US Army Corps of Engineers be prohibitively expensive. Our signature simula-

22 COMPUTING IN SCIENCE & ENGINEERING

 Seismic-Wave Propagation surface-wave propagation through scattering, attenuation,
Scientists have a good understanding of seismic-wave prop- and focusing.2 Unless we understand these variations, we
agation in homogeneous layered media; however, the Earth can’t have realistic expectations for ground sensor systems.3
has heterogeneities at all scales as well as a complex topog- Accurately predicting seismic propagation in regions
raphy unique to any given location. These geologic realities with complex subsurface geology and surface topography
make seismic-wave propagation a complex phenomenon. is a critical step toward interpreting and understanding
Understanding seismic-wave propagation and exploiting it data from these regions. Because we simulate seismic-wave
in intelligent sensor systems is a nontrivial endeavor. Wave propagation attributed to vehicle source types in the pres-
propagation in an elastic medium such as the Earth mani- ence of realistic geological and topographical features, we
fests as compressional P-waves (displacement in direction of use 3D numerical modeling. Advances in computational
propagation), shear S-waves (displacement perpendicular to speed have enabled routine use of 3D finite-difference
propagation), and surface waves. The speeds with which techniques.4,5 As we develop new 3D finite-difference time-
these waves travel differ, as do their paths, resulting in differ- domain (FDTD) techniques and algorithms, we’ll need a
ent arrival times at a given distance for each wave type or validation mechanism to ensure that we’ve properly imple-
phase (S-phase, P-phase, and surface (fundamental and mented the methods. Comparing synthetic data with em-
higher modes)). Further, the phases partially convert to each pirical data is an integral step in verifying these new com-
other as they encounter heterogeneities and discontinuities putational algorithms and methods.
in structure, density, or velocity.
Adding to wave propagation’s complexity is energy atten- References
uation through internal frictional losses (anelasticity), scatter- 1. M. Moran and R. Greenfield, Seismic Detection of Military Operations,
ing, and geometrical spreading with distance. The resulting 97_CEP_511_1, internal report, US Army Maneuver Support Battle
wavefield recorded at a point receiver attributed to a source Lab., Ft. Leonard Wood, Mo.,1997.
at distance is a complex time series of many phase arrivals. 2. S. Ketcham, M. Moran, and R. Greenfield, “The Effect of a Topographic
At the battlefield scale of tens of meters to several kilome- Depression on Guided Seismic Surface Waves,” Proc. Soc. of Exploration
ters, we’re primarily concerned with the seismic surface Geophysicists (SEG) Int’l Exposition and 70th Ann. Meeting, SEG, 2000,
waves generated by moving vehicles and exploiting these pp. 2213–2216.
signals for non-line-of-sight target detection, location, and 3. M. Moran et al., Seismic Feasibility Study in Support of Hornet, final report
discrimination using networked sensors. The seismic surface of Program Manager for the Mines Countermines Demolition Seismic
wave, traveling at 0.88 to 0.95 times the shear-wave veloc- Eagle Team, US Army Armament Research, Development, and Eng. Ctr.
ity, is a relatively slow, high-amplitude seismic phase that (ARDEC), 1998.
propagates at the Earth’s surface, similar to the waves cre- 4. E. Gottschammer and K.B. Olsen, “Accuracy of the Explicit Planar Free-
ated when a pebble is dropped in a pond. Surface waves Surface Boundary Condition Implemented in a Fourth-Order Staggard-
propagating in regions with mild topographies generally Grid Velocity-Stress Finite-Difference Scheme, Bulletin Seismic Soc. Am.,
possess high spatial coherence, show smooth amplitude de- vol. 91, no. 2, 2001, pp. 617–623.
cay as a function of vehicle range, and depend little on se- 5. T. Xu and G.A. McMechan, “Efficient 3D Viscoelastic Modeling with
vere meteorological and seasonal variations.1,2 Local varia- Application to Near-Surface Land Seismic Data,” Geophysics, vol. 63,
tions in topography and geological settings, however, affect no. 2, 1998, pp. 601–612.

tion capability may reduce prototype development a generic tracked vehicle in a realistic geology, we
costs substantively, shorten development times, and generate the geology model, synthesize a source
improve overall system performance and reliabil- signature from a modeled vehicle as vertical forces
ity. In addition, our side-by-side comparisons of at a given time and space, and apply these vertical
empirical and synthetic data illustrate the synthetic forces as input to our FDTD wave-propagation al-
data’s fidelity. This study further validates the data gorithm. Finite-difference seismic simulations
synthesis components—source, propagation algo- over the expected range for battlefield systems re-
rithm, and geological model—and its results illus- quire substantial models and computational dura-
trate the feasibility of simulating seismic signatures tions. Consequently, we used parallel computa-
from battlefield-type sources. tions based on a domain decomposition strategy.
This requires that we perform the computations
Problems and Methodology on multiprocessor computers, such as those avail-
To synthesize the seismic-wave field attributed to able at US Department of Defense High-Perfor-

NOVEMBER/DECEMBER 2004 23
 90 90 90
Frequency (Hz)

Frequency (Hz)

Frequency (Hz)
70 70 70

50 50 50

30 30 30

10 10 10

–8
x10 x10–8 x10–8
2 4 2
Amplitude

Amplitude

Amplitude
2
(m/s)

(m/s)

(m/s)
0 0 0
–2 –2
–4 –2
0 20 40 60 80 100 120 0 10 20 30 40 50 60 0 10 20 30 40 50 60
(a) Time (seconds) (b) Time (seconds) (c) Time (seconds)

Figure 1. Spectrograms using seismic data recorded at the Aberdeen Proving Ground for three vehicle types: (a) M1, (b) T72,
and (c) BMP. The harmonic frequencies are a function of vehicle suspension characteristics and vehicle speed. The uniqueness of
the spectral harmonic lines attributed to a vehicle type let us use this information to discriminate among vehicle types.

Velocity topography and subsurface layering for this region

(meters/second)
of Yuma. There is a slight gradient and roughness
700 to the topography, but no surface features would
significantly disrupt the surface wave’s propaga-
600 tion. Figure 3 shows the geology used for the
study, a 90  660  100-meter (width, length,
depth) subsection of the KGS Yuma study, which
500
Height (meters)

25 we modeled using 0.5-meter node rectangular-grid

20 spacing. The model’s long corridor shape allows
15 400
10
receiver offset distances of 50 meters, while pro-
90
5 viding sufficient room for the vehicle to drive 14
300 seconds (approximately 100 meters) in one direc-
200 60
tion and stay within the model. Models of this ex-
150
200 tent and duration are sufficient for predicting seis-
100 30
mic phenomena at frequencies that are relevant to
East to 50 North to battlefield systems.
west (meters) 0 0 south (meters)
We use a visco-elastic FDTD wave-propagation
algorithm featuring eighth-order staggered-grid
Figure 2. Plotting of the 3D Shear velocity model of the Yuma Proving finite-difference operators.3 To accommodate sur-
Ground Smart Weapons Test Range used in our computations. The face topography, we follow Stig Hestholm and
region’s benign subsurface is characterized by layered sediments with Bent Ruud,4 who incorporated surface topography
only subtle undulations in the subsurface layering. with an appropriate stress-free surface boundary
condition into a visco-elastic FDTD wave-propa-
gation model. This method uses a curvilinear grid
mance Computing Modernization Program transformed into a rectangular computational grid
(HPCMP) centers. of equal grid spacing. We visualize the grid map-
The Kansas Geological Survey (KGS) provided ping by proportionally stretching the rectangular
a subsurface model of Yuma, which we used as our grid vertically so the free surface matches the topo-
geological model for the generic tracked vehicle graphic function. We set the model’s side and bot-
simulations.2 We generally characterize the sub- tom boundaries with a 24-cell-thick absorption
surface using a three-layer medium with subtle un- layer5 to damp out or reduce unwanted reflections
dulations. The surface contour graph of Shear from the sides.
wave velocity in Figure 2 illustrates the mild FDTD output is more versatile than field data,

24 COMPUTING IN SCIENCE & ENGINEERING

as we can choose any number of receivers and re- 600
ceiver configurations for an unlimited number of
scenarios, constrained only by the number of 500
nodes in the grid plane. To increase the synthetic
data’s versatility, we convolve new receiver func- 400

South to north (meters)

R4
tions with the data to simulate a particular sensor’s
corner frequency. To match field data, we convolve 300 R3
a 4.5-Hz geophone’s receiver function with our
synthetic data. 200
R2
A synthesized tracked vehicle serves as the seis-
mic source in our simulation. We generate force 100
inputs attributed to the synthetic tracked vehicle R1
model using dynamic analysis design system
0
(DADS)6 software, determining these force inputs 0 50
from complementary dynamic simulations in- East to west
volving the use of a DADS high-fidelity mechan- (meters)
ical model of the tracked vehicle.7 The DADS
model accurately represents the vehicle’s suspen- Figure 3. The Yuma model portion used in our simulation, with
sion characteristics, including a linked-body rep- exaggerated topography. The arrow represents the vehicle path. The
resentation of each track, while treating the hull geophones are 50 meters from the track and spaced 100 meters apart.
and turret as simple rigid bodies. Track and sus-
pension details account for individual road wheel,
road arm, and track block inertial properties, road
arm and track pin torsional stiffness and damping,
and track tension. Torques applied to the drive
sprockets using a sophisticated controller de-
signed within DADS8 drive the synthetic tank.
The tracked vehicle model has a total of 195 bod-
ies and 356 degrees of freedom. Figure 4 shows
the model’s overall level of detail.
Unconstrained contact force elements with stiff-
ness and damping characteristics determine reac-
tion forces between the individual track blocks and
the vehicle drive sprocket, road wheels, idlers, and Figure 4. The tracked vehicle model. A detailed representation of
support rollers. Similar elements use a modified set suspension elements within the tracked vehicle synthetic source
of Bekker soil equations to determine reaction signature model shows the level of detail used in constructing a
forces between each track block and the ground.9 synthetic tracked vehicle.
The model’s output consists of downward vertical
forces at each tread that’s in contact with the
ground. These downward vertical forces, which we vertical, radial, and transverse seismic compo-
derive from the distributed vertical forces at the ve- nents at all surface model node points. We per-
hicle track/wheel interface, compose the forcing form postprocessing of the resulting files of ve-
history applied in the tracked vehicle FDTD sim- locity amplitude snapshots, and time-series
ulation.10 We apply the vehicle forces to the prop- analysis of waveforms at particular coordinates
agation model at the road wheels rather than the using Matlab. We plot the results and compare
track blocks to avoid aliasing problems resulting them with field data.
from inadequate sampling of the numerous ground
contact elements employed at each track block. For Results
the tracked vehicle, this corresponds to 14 spatially We performed our simulation on an HPCMP 744-
separate inputs at the vehicle’s 14 road wheels. processor grid. The side and bottom absorbing
To generate the seismic data, we had the syn- boundaries provided a reduction of approximately
thesized tracked vehicle traverse the model sur- –25 decibels in reflecting wave particle energy. The
face at 25 kilometers per hour, the average vehi- simulated vehicle travel time was 14.2 seconds, with
cle speed recorded at the Yuma field trials. We time steps of 100 microseconds. The computation
transform vehicle ground forces into synthetic time for the simulation was 96 hours using a T3E.

NOVEMBER/DECEMBER 2004 25
 meter offset. The resulting generic tracked vehicle
simulation yields amplitude and spectral content
600 data similar to observed field data.
Time: 6 seconds In Figure 5, the near-source, high-amplitude
500 waves have swamped out the attenuated waves prop-
agating down the model. We chose the plotting scale
400 to emphasize the propagating wave’s attenuation rate
and coherency. The smooth expanding wave front
300 in the snapshot is characteristic of propagation in a
model with layered geology such as the Yuma
Meters 200 model. Animating several of these snapshots reveals
the continuous vehicle-induced wave propagation
100
and the geology-induced wave diffraction and re-
0
50 fraction. As long as the wave front at a particular fre-
0 quency maintains spatial coherency, we can use ar-
ray-processing techniques to extract the line of
bearing from the seismic-wave field.
Figure 5. A snapshot of the ground surface velocity attributable to the Figure 6a shows a 14-second seismic time series
moving synthetic tracked vehicle. The large-amplitude surface waves and corresponding spectrogram from the tracked
can be used to locate and detect objects out of the line-of-sight and vehicle recorded during a Yuma field test; Figure
carry source-specific frequencies that allow for discrimination. 6b shows the predicted synthetic time series and
spectrogram. The principal waveforms displayed
in the images are high-amplitude surface waves
The resulting output file size for a plane was 5 with cylindrical decay (1/R0.5, R = radius) in the
Gbytes. absence of the topography and geology that might
Our FDTD seismic-propagation code’s output disturb this decay. The surface waves’ approxi-
consists of horizontal, radial, and vertical veloc- mate wavelength is 30±5 meters as the wavefield
ity amplitudes at each node for specified x, y, and moves away from the vehicle.
z model planes. For surface-wave analysis, we In the field data (Figure 6a), the closest point of
look at the vertical velocity component taken at approach is the 14-second tank traverse, and the
the Earth’s surface. Because of the nature of the point at which it passes the receiver is evident in
finite-difference grid, we can use time-series the field data time series and in the spectral en-
analysis at any node or receiver and take snap- ergy from the signals’ maximum amplitude. The
shots of any x, y, or z plane for a wave-field analy- sensor has high-amplitude velocities (approxi-
sis or graphical representation. The particle ve- mately 1  10–4 meters per second) in the first 4
locity snapshot images provide an intuitive seconds of the trace and trails off at the end, indi-
picture of wave-propagation complexity. We can cating that the tank was at or about even with the
also concatenate these snapshots to build movies sensor and then passed it. The spectral amplitudes
that provide a graphical perspective of topo- reflect the same thing and have a notable har-
graphical and heterogeneous wave guide effects monic band at approximately 35 to 38 Hz. Earlier
on the wave propagation at both the surface and work correlates this dominant spectral harmonic
subsurface for complex geologies. with track block spacing and vehicle speed.10
Figure 5 is a snapshot of the vertical particle ve- Lighter spectral bands from 20 to 55 Hz are also
locity w on the model surface for our tracked vehi- visible but lack the intensity of the vehicle’s char-
cle computational run. The easily visible, large-am- acteristic frequency.
plitude surface waves radiating from the source The synthetic time series at the receiver (Fig-
maintain their coherency well past 100 meters. ure 6b) indicates a good match with field data in
amplitude and decay as the vehicle passes the sen-
Synthetic versus Field Data sor at the 50-meter offset. In the frequency do-
We designed the tracked vehicle simulation to main, a dominant spectral harmonic is visible at
model a simple field experiment. In this experi- approximately 35 Hz. The slight disagreement
ment, we specified the locations of the tracked ve- between this and the dominant harmonic ob-
hicle and receiver using GPS, offsetting the re- served in the field data might be attributable to
ceiver 50 meters from the vehicle travel path. We small differences between the simulated and ac-
configured the synthetic case using the same 50- tual vehicle speeds. The other coherent spectral

26 COMPUTING IN SCIENCE & ENGINEERING

Particle velocity (m/s)

Particle velocity (m/s)

–4 Receiver 1: particle velocity history
x10 –4
1 x10 Receiver 1: particle velocity history
1
0
0
–1
–1
110 111 112 113 114 115 116 117 118 119 4 5 6 7 8 9 10 11 12 13 14
–4
x10–4 Particle velocity spectrogram x10 Particle velocity spectrogram
60 60

Particle velocity (m/s)

Frequency (Hz)

50
40 40
30
20 20
10
0 0
110 111 112 113 114 115 116 117 118 119 4 5 6 7 8 9 10 11 12 13 14
(a) Time (seconds) (b) Time (seconds)

Figure 6. The Yuma field test. Time series and spectrogram attributed to (a) a moving tracked vehicle from the Yuma field
experiments and (b) a simulated tracked vehicle traveling over a Yuma geology model. In both models, the receiver is offset 50
meters from the track, and the spectral energy amplitudes from multiple receivers indicate the target direction of travel. The
dominant spectral line at approximately 35 Hz is easily visible in both field and synthetic data.

lines visible in the synthetic spectrogram have less • vertical particle velocity is the quantity a ver-
amplitude than 35 Hz, but are more apparent than tically oriented geophone would record in a
in the real data. The components’ dimensional field measurement, and
consistency in the vehicle dynamics model and the • battlefield seismic sensors will use geophones.
precise control of vehicle speed within the vehicle
model simulations might account for the larger By providing the full wavefield history over the
amount of coherent energy within the secondary model’s surface, synthetic data can support dis-
harmonics in the synthetic spectragrams. Such crimination, location, detection algorithm, and sys-
secondary harmonics might be less likely to ap- tem optimization development.
pear in field data due to normal wear and tear of Researchers or analysts can use high-fidelity sim-
vehicle road wheel and track components, and ulations, such as those we describe, in the develop-
driver- and terrain-induced speed fluctuations. ment of target signature databases, system devel-
Additional tuning of the DADS model might im- opment, and system performance prediction, and
prove spectral matching to field data. can use spectral analysis for target discrimination
and classification. The time delay associated with a
continuous source, such as individual treads on a

Q
uantitative descriptions and predic- tracked vehicle striking the ground sequentially,
tions of the seismic-wave field over causes individual wavefields (attributed to specific
time and space are particularly suited source time) to interfere with each other, intro-
for developing and planning ground- ducing artifacts (such as spectral modulations) that
sensor network deployments because they allow can be diagnostic indicators of the source type. Sci-
placement of virtual sensors at any point in the entists in the nuclear monitoring community com-
simulation domain. Most ground-sensor system monly use this technique to discriminate the ripple
functions include target detection and system fire of mining explosions from that of impulsive
wake-up, target range and bearing estimation, and sources generated from a single large buried ex-
target classification. These features, combined plosion.11 Targets typically found in the battlefield
with a signature discrimination capability, provide environment, such as moving personnel or vehi-
for near-real-time battlefield situational awareness. cles, also generate a continuous seismic source that
Such simulations have obvious utility because we can identify using spectral content in the
ground sensor data.
• vehicles generate unique spectral characteris- Our comparison of moving vehicle field data and
tics that can be replicated, simulations is the first such validation of vehicle sim-

NOVEMBER/DECEMBER 2004 27
 ulations produced by a seismic model. The synthetic 6. DADS Overview, LMS Int’l, pub. no. 4.0/2090/A20/11.99, Nov.
wavefield particle velocity images are good matches 1999.

in both amplitude and spectra, as predicted from 7. J. Lacombe, M. Moran, and S. Happel, “A 3-Dimensional Dy-
namics Model For Generating Tracked Vehicle Seismic Signals,”
seismic propagation and physics. The sensor exam- Proc. 2000 Meeting of the Military Sensing Symposia (MSS), spe-
ined matches the principal features of the seismic cialty group on battlefield acoustic and seismic sensing, report
signatures observed in field measurements, making 440000-195-X (I), Infrared Information Analysis Center, Veridian
Systems Division, 2001, pp. 377–383.
target discrimination and identification a very real
8. J. Lacombe, M. Moran, and D. Harjes, “Mechanical System Sim-
possibility. The vehicle-dependent frequency con- ulations for Seismic Signature Modeling,” Proc. 2001 Meeting of
tent in seismic signals and the ability to measure it the Military Sensing Symposia (MSS), specialty group on battle-
in the field suggest that we could construct large li- field acoustic and seismic sensing, Infrared Information Analysis
Center, Veridian Systems Division, 2001.
braries as a tool for real-time discrimination. Be-
9. M.G. Bekker, Introduction to Terrain-Vehicle Systems, Univ. of
cause we can synthetically generate vehicle signa- Michigan Press, 1969.
tures, we could generate signatures of unobtainable 10. S.A. Ketcham et al., “Modeling Ground Loading by Moving
foreign and concept vehicles, as well as supplement Tracked Vehicles in FDTD Seismic Simulations,” Proc. 2002 Mili-
empirical data sets acquired from field trials. tary Sensing Symposia (MSS), specialty group on battlefield
acoustic and seismic sensing, Infrared Information Analysis Cen-
More generally, these results indicate that simu- ter, Veridian Systems Division, 2002.
lations have an important role in seismic system de- 11. M.A. Hedlin, “An Analysis of Low-Frequency Seismic Signals
velopment and acquisition. They can reduce sys- From Large Mining Blasts,” Proc. 21st Seismic Research Symp.:
tem costs and development time, improve system Technologies for Monitoring the Comprehensive Nuclear Test Ban
Treaty, LA-UR-99-4700, 1999, pp. 713–722.
performance in complex environments through re-
ceiver placement optimization planning, and allow
propagation physics to be incorporated into system Thomas S. Anderson is a research physical scientist at
detection and discrimination algorithms. the US Army Corps of Engineers, Engineer Research and
We used seismic data synthetic production for Development Center, Cold Regions Research and Engi-
M1, T72, and BMP vehicle targets with similar neering Laboratory (ERDC-CRREL). His research inter-
success. ests include 3D seismic-wave propagation and the ef-
fects on detection, battlefield awareness, and ordnance
Acknowledgments location using seismic-acoustic arrays. Anderson has a
The US Army Office of the Program Manager for PhD in Earth sciences with a specialty in regional seis-
Mines, Countermines, and Demolitions supported this mic propagation. Contact him at [Link]@
work, with the US Army Engineer Research and [Link].
Development Center, Cold Regions Research and
Engineering Laboratory providing additional funding Mark L. Moran is research geophysicist at ERDC-
under work item PE62784/AT42. The US Department CRREL. His current interests include seismic and
of Defense High-Performance Computing Modern- acoustic research and development of networked
ization Program under Challenge Project C-76 ground sensor systems and battlefield awareness ef-
provided computational resources. forts. Moran has a PhD in geophysics. Contact him at
[Link]@[Link].
References
1. S.A. Ketcham et al., “Soil Attenuation in Seismic Simulations; Im- Stephen A. Ketcham is a research physical scientist at
plications for Vehicle Tracking,” Proc. 2001 Meeting of the Mili-
tary Sensing Symposia (MSS), specialty group on battlefield ERDC-CRREL. His research interests include seismic and
acoustic and seismic sensing, magnetic and electric field sensors, acoustic finite-difference modeling in support of battle-
CD-ROM, Infrared Information Analysis Center, Veridian Systems field awareness efforts. Ketcham has a PhD in civil engi-
Division, 2001.
neering. He is a member of the IEEE and the American
2. R.D. Miller et al., 3D Characterization of Seismic Properties at the
Smart Weapons Test Range, YPG, report no. A913904, USACE Society of Civil Engineers. Contact him at steven.a.
Cold Regions Research and Eng. Lab, Oct. 2001. ketcham@[Link].
3. M. Moran, S. Ketcham, and R. Greenfield, “Three-Dimensional
Finite-Difference Seismic Signal Propagation,” Proc. 1999 Meet- James Lacombe is a research mechanical engineer
ing of the Military Sensing Symposia (MSS), specialty group on bat-
tlefield acoustic and seismic sensing, report 440000_147_X, at ERDC-CRREL. His research interests include com-
ERIM Int’l, 1999, pp. 1–12. puter model development for predicting wheeled
4. S. Hestholm and B. Ruud, “3D Finite-Difference Elastic Wave and tracked vehicle seismic signatures, environmen-
Modeling Including Surface Topography,” Geophysics, vol. 63, tal effects on target thermal signatures, vehicle mo-
no. 2, 1998, pp. 613–622.
bility, and electro-optical sensor performance. La-
5. C. Cerjan, R. Kosloff, and M. Reshef, “A Nonreflecting Boundary
Condition for Discrete Acoustic-Wave and Elastic-Wave Equa- combe has a BS in mechanical engineering. Contact
tions,” Geophysics, vol. 50, 1985, pp. 705–708. him at [Link]@[Link].

28 COMPUTING IN SCIENCE & ENGINEERING