Sie sind auf Seite 1von 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/268262430

COMBINING GEODESY WITH SEISMOLOGY IN THE NEW MADRID SEISMIC ZONE

Article

CITATIONS READS
0 75

1 author:

Noah Lenstra
University of North Carolina at Greensboro
19 PUBLICATIONS   14 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Movement-Based Programming in Public Libraries View project

All content following this page was uploaded by Noah Lenstra on 06 April 2016.

The user has requested enhancement of the downloaded file.


COMBINING GEODESY WITH SEISMOLOGY IN THE NEW MADRID SEISMIC
ZONE

Noah Lenstra
Eastern Illinois University
Center for Earthquake Research and Information (CERI) at the University of Memphis
Professor Robert Smalley, Jr.
Abstract:

As GPS technology develops, increasingly it is used to measure movement across active faults.
However, geodetic research still remains separate from traditional seismology in the study of
earthquakes. Following the December 26 Sumatra earthquakes, the surface waves were large
enough to be seen using GPS in the eastern United States. Taking compiled GPS and seismic
data from across the central and eastern United States; we use established techniques to render
relative movement across a number of continually collecting stations. A comparison gives a
first positive correlation between the GPS and seismic data. In addition, we apply surface wave
analysis techniques to both data types to see if they share phase and group velocities, and hence,
dispersion curves. A problem in a correlation is that few locations contain both GPS and seis-
mic stations; hence geographical nearness must be taken into account while processing the data.
If the correlation can be made—despite errors—between the GPS data and seismograms, two
important, but disparate, techniques could be united to better understand our knowledge of
ground deformation resulting from earthquakes.

Introduction

The Center for Earthquake Research and Information (CERI) at the University of Mem-
phis maintains both a continually operating seismic and GPS network around the New Madrid
Seismic Zone. However, the two networks are operated completely independent of each other.
By combining the two technologies, GPS, which is becoming increasingly popular in the civil-
ian world for navigation, could be made an important complement to seismology in areas that
do not feature seismic networks due to cost, and/or feasibility restrictions. As GPS technology
continues to improve, the technology could prove vital in the observation of future earthquakes
around the world (Smalley, 2005).

The reason that no attempt has been made to combine GPS and seismology is that few
people have tried to use the same types of filtering and analysis protocol for both technologies.
In traditional surface wave analysis, a number of steps are performed. First, any response con-
volved with the signal due to the instrument recording is attempted to be removed. Next, dis-
persion curves are created for both group and phase velocity. Finally, the dispersion curves are
inverted relative to an earth model to see changes in seismic velocity with depth, and thus to see
changes in earth composition with depth. This technique has been used with great success for
many years in seismology (Ketter, 2002), but has not been utilized with GPS technology.

There are many reasons why it is not trivial to apply surface wave analysis to GPS.
GPS and seismology record fundamentally different phenomena, making unification difficult.
GPS, a geodetic technology, relies upon a number of continually orbiting satellites run by the
United States military to record relative position, both vertically and laterally, relative to a geo-
detic earth model. Seismology, on the contrary, does not rely upon any such complicated setup;
in its simplest form, seismology merely records earth "shaking" by means of inertia, leaving a
concrete picture of surface waves without the complicated signal processing necessary in GPS
to see such waves. Finally making the unification unlikely in the past, numerous errors that
made GPS a very inexact science, among them: satellite noise, ionosphere disturbances, and
secondary earth movements not directly related to earthquakes.

Sumatra-Andaman earthquake enables GPS-seismic Correlation

On December 26, however, an extraordinary event, in terms of earthquakes, created a


situation where the errors could be disregarded due to the large size of surface waves generated
from the Sumatra-Andaman
(S-A) Magnitude 9.0 earth-
quake (fig. 1). When an
earthquake occurs, in addi-
tion to body waves that
travel through the earth, two
types of surface waves,
Love and Rayleigh, are gen-
erated that travel over the
surface of the earth. The
surface waves from the S-A
earthquake traveled across
the Indian Ocean, Asia, Si-
beria, the North Pole, and
Canada, before arriving in
the New Madrid Seismic
Zone roughly 50 minutes
after the earthquake oc-
Fig. 1— Map of earthquake surface wave path from the epicenter curred at 00:58:53 (UTC).
to the CERI broadband network. Although in most cases such
a long great circle path
would mean that the waves would be attenuated to the point of invisibility to GPS technology,
the large size of the original earthquake means that the wave pattern is clearly visible in the
GPS record. Thus, in this paper I will illustrate the commencement of an ongoing process to
combine geodesy and seismology through surface wave analysis.

Data collection for seismic-GPS analysis

The long-term goals of this project are not only to integrate GPS and seismology in the
New Madrid seismic zone, but across the Eastern United States. Thus, the first step was to col-
lect broadband seismic data and continuously collected GPS data from across eastern North
America. To do this, we contacted representatives from GPS networks and seismic networks,
as well as utilized the interactive, online seismic data retrieval site, IRIS DMC (fig. 2) We
choose broadband seismic data, as opposed to short period and many other types, so that we can
examine a broad range of periods in our analysis. Similarly, 1-second GPS data (or GPS sites
that record new coordinates every 1 second) was chosen because this is currently the fasting
operating type of GPS site.
Fig. 2 — Map of Eastern US, data to be used in seismology-GPS studies. Red and purple (dark)
triangles indicate GPS stations and yellow triangles (light) indicate broadband seismic stations.
Maps created in GMT.

First order correlation

With data in hand, we attempted to make a first-order correlation between the GPS and
seismic records by using TRACK, a GPS processing program in the GAMIT/GLOBK software
package, to hold one station "fixed" and allow up to four other GPS stations to move relative to
the "fixed" station, generating a kinematic picture analogous to a seismic record. In tandem, we
converted a velocity seismogram to a relative displacement seismogram by removing the instru-
ment response, and subtracting one seismogram from another. As Figures 3 and 4 shows, the
technique proved extremely successful in proving that a first order correlation between GPS and
seismology is a not too difficulty obtained goal. In this first-order correlation, we took two GPS
Fig. 3, 4 Above, right, the
close correlation between
the seismic data and GPS
data can be seen in these
Matlab plots. Note that
the units of the y-axis are
in meters, and not centi-
meters, as is listed.

sites and two broadband


seismic sites that were very
near one another. However,
there was a distance of ap-
proximately 10 miles be-
tween one of the axes of
the correlation, signifying
that if the GPS and seismic
stations were literally one
upon another, instead of
geographically separated, we would expect an even closer correlation in our data.
About Surface Wave Analysis

After finding the surface waves (in this study, we choose to observe Love waves, but the
techniques could be applied with equal success to Rayleigh waves), we began implementing
standard surface wave analysis techniques through the utilization of Robert B. Herrmann's
"Computer Programs in Seismology" (Herrmann, 2002). We choose to do the seismic side of
the analysis first, due to its better accuracy, so that we would have a benchmark for future stud-
ies. In addition, rather than use the “straingrams” generated in the above example, we utilized
normal seismograms. The reason being that we are not sure if “straingrams” will invert for earth
structure in the same manner as simple seismograms. More research will have to be performed
to address this problem. In the mean time, we do the standard seismology to establish parame-

Fig. 5—Display of the three original SAC files, windowed to see just the surface waves, and the
two rotated component seismograms. From top, the original East-West seismogram, the rotated
transverse seismogram, the rotated radial seismogram, the original North-South seismogram,
and the original vertical seismogram. Note the strong presence of the Love waves in the East-
West and transverse seismograms. Also notice how little a shift occurs between East-West and
transverse and North-South and radial. By definition, the transverse component of a seismogram
is that which is perpendicular to the path of the surface waves and the radial is that which is par-
allel to the path of propagation.
ters within which our GPS results should lie, similar to the preparation of a “straingram” to
compare to the GPS displacement in the above example.

Preparing the Seismograms for Surface Wave Analysis

The first step in any seismic surface wave analysis is to remove the instrument response
so that the analysis is done only to how the earth filters the surface waves, and not how the in-
strument itself may have filtered and dispersed the seismic signal. In this study we utilized both
instrument response removal techniques of analyzing poles and zeroes and also any phase and
amplitude shifts that may be due to the seismometer, via the program evalresp. In addition, fre-
quency limits were chosen to limit the area of future study while removing the instrument re-
sponse, the frequency limits being chosen so as not to extend beyond the niquist frequency, the
limiting frequency that a seismometer can pick up (Kovach, 1978).

The next step is to rotate the seismograms to produce radial and transverse components
from the initial north/south, east/west, and vertical components. This was done by calculating
the back azimuth and great circle path from the event latitude and longitude, and station(s) lati-
tude and longitude. Due to the path of the surface waves, the back azimuth was nearly naturally
polarized (that is equal to 0/360 degrees), so little rotation was necessary before beginning
analysis (fig. 5).

When the instrument response is removed and the data rotated, we cut the files so that
the dispersion curves display only the dispersion curves of the wave types we are interested in.
In reality, it is nearly impossible to completely remove traces from other waves, but the nearest
approximation is attempted. Since we are looking for Love waves, we assumed speeds of be-
tween 3 and 5 km/s from the earthquake hypocenter. Our stations average a distance of 15000
km from the earthquake epicenter, so we cut our data in the window from 50 minutes and 83
minutes after the earthquake occurred (Oliver, 1962).

Dispersion Curve Analysis

Since in the program we used, group velocity depends only on one seismogram, whereas
phase velocity depends on multiple seismograms spread out over a geographic area along a
great circle arc, I opted to perform the group velocity analysis first. Group velocity, put simply
is the speed of a wave packet, rather than an individual wave. Mathematically it is change in
angular velocity over the change in the wave number. Phase velocity, on the other hand is the
speed of the individual phases inside the wave packet. For these reasons, phase velocity is rep-
resented by angular velocity over the wave number, for a particular phase, and not a group of
phases traveling in tandem as in group velocity.

The phase and group velocities are not constant for each frequency and wavelength, but
depend on the materials through which they pass before arriving at the recording stations. Since
long wavelengths extend deeper into the earth, they touch more dense material, and hence,
Fig. 6—Output of Robert Herrmann’s sacmft96 program, displaying the group velocity inversion
data for CERI seismic station GLAT, in the New Madrid Seismic Zone.

move more quicker, than the short wavelength waves that only travel through the near-surface
material. Hence the long-wavelength waves are dispersed and arrive before the short-
wavelength waves (van der Lee, 2002). The amount of this dispersion can be used through it-
erative inversions to create a non-unique picture of the earth structure beneath a particular re-
gion, an important analysis for finding out, for example, what the fundamental harmonic is of a
landmass such as the Mississippi embayment, a critical piece of knowledge for engineers and
city planners.

In any case, using Herrmann’s program sacmft96 we generated group velocity disper-
sion curves for a number of broadband seismic stations, paying particular attention to the New
Madrid Seismic Zone. When the group velocity plot is generated, one must manually select the
smooth-fit curves that appear to represent linear changes in group velocity, this being the curves
most likely to yield constructive inversion information (fig. 6).

In this program the exact group velocity dispersion curve is generated by the program’s
applying a Gaussian filter to the seismogram around frequencies selected by the user. This in
turn produces a number of clearly visible maximum amplitudes, which correlate to the arrival of
the group of waves. Since we thus know the travel time of the group, and we know the distance
from the hypocenter, group velocity can be easily calculated. However, complicating the calcu-
lation is the choice of alpha, the filter parameter, which is a non-trivial choice. Alpha should
increase with distance from the hypocenter, and becomes increasingly unstable with distance
from the earthquake. Since we are looking for dispersion curves at a very high period (above
100 seconds) and also a great distance from the earthquake (over 15000 km), our choice of al-
pha could greatly compromise our dispersion curves, something which may not become evident
until further research is carried out.

Phase Velocity Dispersion Curve

After the group velocity dispersion curves are generated, we take the same seismograms used to
calculate the group velocity and use them to cal-
culate the phase velocity in a program called
sacpom96. In this program, for each frequency the
program discovers when each particular phase
arrives, and compares the arrival at one station
with the arrival at other stations along the great
circle path to determine the speed of the phase in
a certain geographic region. For our study we
took stations from St. Louis, Missouri to Oxford,
Mississippi, in order to see long wavelength
waves that require a broad geographic spread
(Fig. 7).

With the stations chosen, and the seismo-


grams prepared, we perform the phase velocity
dispersion curve analysis with a number of rays
chosen at 100. Rays, in sacpom96 are analogous
to the alpha in sacmft96, that is, they determine
the filter parameters that will be used in the analy-
sis. I choose 100, because any more would need-
lessly complicate the data and any less would
make the trends in the dispersion curve too ob-
scure to see (Fig. 8).

When the dispersion curves are generated and


chosen, the next logical step is to invert for earth
Fig. 7—For the phase velocity dispersion
curve, I drew upon the broadband seismic structure. By inverting for earth structure, we can
stations at left. The most north station, SLM, see into the surface of the earth and learn valu-
is in St. Louis, Missouri, and most south sta- able insights into earthquake risk important to
tion, OXF, is in Oxford, Mississippi. engineering.

For the best inversions, it is important to combine phase and group velocities for both
Love and Rayleigh waves with receiver function inversion. However, since receiver function
inversion requires the correlation of P-wave arrival times at multiple stations, it could not be
performed since at these stations the P-waves are not visible in the seismograms due to the de-
gree distance from the earthquake. For this reason, we performed merely a preliminary inver-
sion in surf96, another pro-
gram in Herrmann’s software
suite, with the concatenated
dispersion curves. A very pre-
liminary inversion can be
seen in Figure 9.

In the inversion pro-


gram, a single dispersion
curve is needed. Since we
generated multiple dispersion
curves, we therefore needed
to sum and average the data
for each frequency/period and
delete any values that appear
completely contrary to the
trends. By finding values that
do not seem reasonable, we
can in turn go back to original
dispersion curves and see if
they are reasonable. In this
manner there is a constant
give-and-take between the
Fig. 8—Phase velocity dispersion curve of the geographic array dispersion curves and the in-
on the previous page. The lack of linearity in the curve may be version as the two are simul-
due to many things, such as multi-path, the earthquake waves not taneously improved to give
strictly traveling on the great circle arc, but going on different
paths due to lateral heterogeneities, errors in the respective seis- an accurate estimation of the
mograms, and other random error. To a large degree, phase ve- structure beneath the study
locity is more problematic than group velocity since it relies on area. The full inversion will
multiple stations, rather than just one station. be produced at a later data
with much more measure-
ment and analysis. At this point, it is enough to show that a preliminary inversion is possible
from the few data that we have thus collected and dispersion curves that we have generated.

Conclusions and Further Work

The work begun in this study will be elaborated upon in the future by students at the
Center for Earthquake Research and Information (CERI). The first step will be to take the rela-
tive displacement data from the GPS and make it into a seismogram that can be read by SAC,
and thus can be read in the suite of Herrmann’s programs used in this study. The next step is to
see if relative displacement GPS inversions can yield relevant inversions. Relative displacement
GPS data and “straingrams” are not the same as regular seismograms. Simple notions of group
and phase velocity become problematic due to the constructive and destructive trends that occur
when two stations are moving simultaneously, due to seeing the same earthquake waves and
having only the difference between these motions recorded. Further theoretical research and
experimentation must be performed to see if the relative movement of the GPS data can be used
to perform the same type of surface wave analysis begun in this study. In any case, the work
done here should prove helpful as a cornerstone for future work in combining GPS and seismol-
ogy in the study of earthquakes.

Fig. 9—Example of dispersion curve inversion. Note that for the final version the dispersion ele-
ments must be averaged for each frequency, so that only one velocity exists for each period, and
an earth model of the New Madrid Seismic Area must be produced to check the data.
Appendices: Further Images of Surface Wave Analysis

Preliminary
“straingram” group
velocity dispersion
curve from the seismo-
gram seen in Figures 3
and 4. This group ve-
locity dispersion curve
is much more problem-
atic than in the case of
the simple one station
seismometer. In addi-
tion, the physical sig-
nificance is compro-
mised to the presence
of the group arriving at
both stations, yet the
program is looking for
the arrival at a single
station. More work
must be done to see
what “straingram” dis-
persion curves signify.

At the top is the origi-


nal GLAT transverse
component seismo-
gram cut to the win-
dow of interest. Be-
neath this are the two
SAC files produced
from analyzing the
group velocity in
sacmft96. The middle
seismogram repre-
sents the idealized
seismogram if the sur-
face waves traveled
solely along the great
circle path without any
straying. Below are
any residual motions
due to lateral hetero-
geneities that cause
the surface waves to
arrive at times other
than that predicted by
the surface wave
equation. The two seismograms added together will produce the top seismogram, the one actu-
ally recorded by the broadband instrumentation.
Works Cited:

Herrmann, Robert B. JV Computer Programs in Seismology St. Louis University, St. Louis, Ill.
2002

Ketter, Brett “Seismic Velocity Structure of Central Asia from Surface-Wave Dispersion” Un
published Master’s Thesis. St. Louis University

Kovach, Robert L. “Seismic Surface Waves and Crustal and Upper Mantle Structure” Reviews
of Geophysics and Space Physics 1978; 16(1): 1-13

Oliver, Jack “A Summary of Observed Seismic Surface Wave Dispersion” Bulletin of the Seis
mological Society of America 1962; 52(1): 81-86

Smalley, Jr., Robert et.al. “Space geodetic evidence for rapid strain rates in the New Madrid
seismic zone of central USA” Nature 2005 June 22

Van der Lee, Suzan “High-Resolution estimates of lithospheric thickness from Missouri to
Massachusetts, USA” Earth and Planetary Science Letters 2002; 203: 15-23

View publication stats

Das könnte Ihnen auch gefallen