Académique Documents
Professionnel Documents
Culture Documents
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/226662178
CITATIONS
READS
78
562
1 author:
T. A. Herring
Massachusetts Institute of Technology
173 PUBLICATIONS 9,083 CITATIONS
SEE PROFILE
Abstract Over the past decade, many Global Positioning System (GPS) networks have been installed to
monitor tectonic motions around the world. Some of
these networks contain hundreds of sites spread
across active tectonic margins where the differences
in velocities across the network can be 50100 mm/
year. For networks that have been running for a
number of years, the uncertainty in the velocity
estimates can be less than 1 mm/year. In some cases
the vertical motions can also be significant and of
importance. Often, the time series of the motions of
the GPS sites show complex non-linear behavior, and
in all cases the statistical model of the time series is
more complex than simple white noise. In this article,
we describe a set of Matlab tools developed for use
with the GAMIT/GLOBK GPS data analysis system
(King 2002; King and Herring 2002) that allow interactive viewing and manipulation of GPS velocities and
time series with a Matlab-based graphical user
interface (GUI). The formats of the data files used by
the tools are specific to GAMIT/GLOBK, but they are
simple ASCII files that can be generated from other
file formats. The tools are referred to as GGMatlab.
194
Introduction
The generation of tectonic velocity fields with large numbers
of GPS sites and complicated time series is becoming common. The complications in the time series are often associated with geophysical effects such as offsets due to
earthquakes, postseismic transient behavior after earthquakes, and more noise-like phenomena such as the effects
of groundwater changes. Complications in time series can
also arise from offsets due to GPS antenna and receiver
changes and the addition or removal of antenna raydomes
from a GPS site. In some cases, GPS receivers fail in modes
that generate reasonable carrier-phase measurements but
corrupt position estimates. The GGMatlab tools are designed
to allow both the generators and users of large GPS velocity
fields to examine in detail the velocity fields and the time
series from which the velocities were generated. The tools are
documented and available from http://www-gpsg.mit.edu/
tah/GGMatlab. The current version is 1.02 but the tools are
still being developed and future versions are expected.
The GGMatlab toolbox currently contains two main
components: velview, which allows viewing and analysis of
velocity fields; and tsview, which allows viewing and
manipulation of time series. The time-series viewing tool
can be invoked from within velview by clicking on a
plotted velocity vector. Within velview, profiles of velocities across regions can be generated and functions fit to the
velocity changes along the profiles. Velview allows two
different velocity field files to be overlain and the statistics
of the differences and biases between them calculated.
All of the details of the options and installation of
GGMatlab are given on the Web page. Here, we describe
some of main features and the typical scenarios for using
the packages. The package is designed to work with Matlab
release 12 and greater. Linux executable versions (kernel
version 2.4.7 and greater) are also available. The executable versions do not support all the features available when
the package is run from within Matlab.
Velview
The primary aim of velview is to allow plotting, manipulation and comparison of velocity fields generated by
GLOBK or other programs if the appropriate format files
are generated. An example of a velview screen view is
shown in Fig. 1 with some of the features invoked. The
example shown is taken from the combined analysis of the
DOI 10.1007/s10291-003-0068-0
Fig. 1
Example of a velview screen image. In this example, two velocity fields
have been loaded from the SCIGN analysis. The primary field is
displayed in red, and the secondary in blue. The Feature File option
has been used to add the California coast (black), the major faults in
the area (green lines) and the sesismicity (brown dots, earthquake
magnitudes greater than 2 since 1996). The velocity vector of
LDES_CHT has been labeled on the figure in a yellow box (box will be
removed if it is clicked on), and the vector options are shown for the
site ISLK_GHT (grey box; the grey box will disappear once an option
is selected or the site name clicked.)
195
program. If the time series files are not in the same directory
from which velview was run, the name of the directory can
be specified in the box next to the Time Series label at the
top of the figure. Tsview is discussed below.
Another feature of velview is the ability to select a transect
across the velocity field for profiling where the velocities are
shown as a function of distance along the profile and are
projected into directions based on the azimuth given in the
box next to the AZ button. The azimuth of the projection
can be either typed into box, graphically selected by clicking
the AZ button when the box is empty, or default to along
the profile direction and normal to the profile direction. The
profile is selected by first drawing the profile centerline and
then by graphical selecting its width. The selected profile
box is displayed on the figure and the sites inside the box are
marked with yellow circles. The profile box information is
placed on a stack as profiles are selected. The boxes and site
markings on the figure can be removed from the top of the
stack using the POP button (most recent removed) or
from the bottom of the stack (earliest removed) using the
PIP button. Only results from the primary velocity field
are shown on the profiles. (The SWAP button at the bottom of the velview window can be used to exchange the
velocity fields thus allowing profiles for each field to be
plotted in separate windows). An example of the profile plot
is shown in Fig. 2. Part (a) of the figure shows the velview
window used to select the profile and part (b) shows the
profile figure itself. For this figure, the azimuth for the
projection of the velocities along the profile was graphically
selected to be parallel to the CICE velocity vector.
In Fig. 2b, we have used the Fitting Functions window to fit
a function to the profile velocities based on the co-seismic
strain accumulation for a locked strike-slip fault (the SS
function) (Okada 1985). When velview is run in Matlab, any
valid Matlab function, defined to be a function of distance
along the profile, can be used in the fitting functions. In the
executable version, only specific functions can be used with
the pre-defined SS function being available in both the
Matlab and executable versions. The green shaded boxes on
the left of the figure show the results of the fit in the direction
of the azimuth (labeled along track), normal to this
direction (labeled cross track) and for the vertical velocity.
Because space is limited in these text boxes, the units are not
given. For the weighted root mean square (WRMS) scatter
and the amplitudes of the estimated coefficients (FCN0 and
FCN1 in the figure), they are in mm/year. The FCN0 value is
the offset of the profile velocities. The normalized rootmean-square (NRMS) scatter is dimensionless and is the
ratio of the scatter to the expected scatter based on the
velocity error bars (square root of v2-per-degree-of-freedom). A single fault with a deep locking depth has been used
to fit the profile in the example but a better fit can be obtained
by fitting to multiple faults with shallower locking depths.
Individual sites may be deleted before the fitting functions
are estimated and they appear with a cross over the point.
The results of the fit and the profile information can be
saved to a file by putting the file name in the yellow File
box and clicking the save button. Information may be
extracted from this file so that it can be plotted in another
graphics program if desired.
Fig. 2
a The velview window used to select the profile sites shown in Fig. 2b.
The solid straight-line shows the center of the profile and the dotted
lines mark the region of the sites to be included in the profile. The
profile starts in the northeast and ends in the southwest. b The profile
window generated by velview when the profile button is selected. The
SS function in the Fitting Functions models the behavior of an
infinite strike-slip fault. In the example, the fault is locked to a depth
of 20 km and crosses the profile 200 km from the start of the profile.
The blue line in each frame shows the model values and the red line in
each frame is the mean value of the velocities. The green boxes on the
left show the results of the fit and can be removed from the figure by
clicking on them. All error bars are one standard deviation
Tsview
Tsview is the time series viewing and manipulation module
in the GGMatlab toolbox. Tsview can be invoked by selecting TimeSeries in the popup menu that appears when a
velocity vector is clicked in velview or it can be run as a
stand-alone module. Tsview plots times series of GLOBK
north, east and height (NEU) coordinate estimates
contained in GLOBK time series files. The north coordinates
are the geodetic latitudes of the sites multiplied by the
WGS84 semi major axis, the east coordinates are the
distances from the Greenwich meridian along the small
circles at (quantized) latitudes of the sites, and the heights
are WGS84 ellipsoidal heights. The latitudes of the sites are
quantized for the east coordinate calculation so that the east
197
Fig. 2 b
(Contd.)
198
the time series and the statistical properties of the time series
residuals with either white or correlated noise assumptions.
If the RealSigma (meaning realistic sigma) button is
unchecked, a white noise model is used. The white noise
assumption almost invariably generates very optimistic
estimates of the uncertainties of the parameter estimates.
When the RealSigma box is checked, a time-correlated noise
model is used to estimate the parameter uncertainties. With
this model, a correlation time of the residuals for each
coordinate component is estimated by computing the increase in the chi-squared-per-degree of freedom of successively longer time averages of the residuals. For a white noise
error model, the chi-squared-per-degree of freedom would
not depend on averaging time. With temporal correlations
in the time series, chi-squared-per-degree of freedom
increases as the residuals are averaged over successively
longer time intervals. The character of the averaged residual
can be viewed in tsview using the Average button. The
value in the box below the button gives the length of time in
days that will be averaged. The statistics of the averaged
residuals are shown in bold on the display. For the example
shown in Fig. 3, the realistic sigma estimates of the secular
velocity in each coordinate of this site are 3 to 5 times larger
than the white noise estimates. Thirty-day averages of the
coordinate residuals have scatters of 0.4 mm in the
horizontal components and 0.8 mm for the vertical
component. These scatters are only a factor 2 smaller for
horizontal coordinates and a factor 4 smaller for the vertical
coordinate than the scatters of daily values. If the residuals
were white noise, the scatters should have reduced by a
factor of 5.5 (square root of 30).
Fig. 3
Example of a tsview screen image. The time series shown is for the site
DHLG which is one of the many sites in Southern California affected
by the October 1999 Hector Mine earthquake (Hudnut et al. 2002). In
the GLOBK processing, this site is divided into a pre-Hector mine
name (DHLG_GLA) and a post-Hector mine name (DHLG_GHT).
The plot shows positions from both site names appended together
with a break in late 1999 (1999/10/16). A logarithmic function is used
to model the behavior of the site position after the earthquake. The
green horizontal lines show the bounds of 3-times the WRMS scatter
of the residuals. The red, nearly horizontal lines near zero on the plots
show the 1-r limits of the model time series based on the standard
deviations of the parameter estimates with the contribution from the
overall mean value of the time series removed. (When only a rate is
estimated, this display forms a thin x cross shape that crosses zero
near the center of the plot.) The full set of parameter estimates and
their standard deviations are written to the Matlab workspace and to a
pop-up window in the upper right hand corner of the computer
screen. In this figure we have used the Realistic Sigma model
discussed in the text
they are used.) The edit file from tsview can be used in
GLOBK to remove edited sites from an analysis and to
implement discontinuities in time series through renaming
sites. The lowest box in this part of the figure allows an output
file from the GLOBK ensum program to be used to sort the
list of sites based on the statistics of their fits to linear motions with the poorest fit sites appearing at the top of the list.
References
Hudnut KW, King NE, Galetzka JE, Stark KF, Behr JA, Aspiotes A,
van Wyk S, Moffitt R, Dockter S, Wyatt F (2002) Continuous
GPS observations of postseismic deformation following the 16
October 1999 Hector Mine, California, earthquake (Mw 7.1),
Bull Seis Soc Am vol 92, no 4, pp 14031422
King RW (2002) Documentation for the GAMIT GPS analysis
software, MIT Internal Report, 206 pp (http://www-gpsg.mit.For users of GAMIT/GLOBK, the results from tsview can be edu/simon/gtgk/GAMIT.pdf)
King RW, Herring TA (2002) Global Kalman filter VLBI and GPS
used in later GLOBK analyses of the data set. The Save
analysis program, MIT Internal Report, 98 pp (http://wwwbutton in center column will save the results from the time
gpsg.mit.edu/simon/gtgk/GLOBK.pdf)
series being displayed in the files whose names are given in
Okada Y (1985) Surface deformation due to shear and tensile
the bottom part of the left-hand column. (The small W
faults in a half space. Bull Seismol Soc Am 75:11351154
buttons show the files to be written, and the E buttons
Watkins MM, Bock Y, Hudnut KW, Prescott WH (1997) The
indicate which files will be erased before they are written. The Southern California Integrated GPS Network: Status Report, Eos
Trans AGU, AGU Spring Meeting, 29 April 1997, p 105
erase buttons reset automatically to non-erase mode once
199