SCIGN MIT Report

March 26, 2003

Thomas Herring, MIT contribution

Last Updated Thursday, March 27, 2003

Combined SIO/JPL Velocity solution

This report updates the Dec 2002 report on the comparison of the JPL and SIO analyses of the SCIGN GPS data collected between Jan 1996 and Oct 2002. In this report we extend the previous comparisons to generate motion models from the combined SIO and JPL analysis of the SCIGN data. We also include is this distribution, time series from the processing of SIO and JPL SINEX files, and from the analyses carried out at each center. All this files have been brought to a common format for both the velocity fields and the time series, and these files can be viewed with the GAMIT/GLOBK Matlab toolkit distributed as GGMatlab.

The following files and directories are included in the distribution SCIGN_COM.tar.Z (Compressed this is a 63Mbyte file, uncompressed it is 270Mbtes).

File/Directory	Description
COM_96_0210.vel	Combined JPL/SIO velocity field for the sites that show small non-secular motions. Generated by GLOBK with site-specific process noise for all sites and additional process noise plus offsets associated with earthquakes.
COM_96_0210_RS.vel	Time series analysis with realistic sigmas based on statistical behavior of time series with successive averaging of time series. For post Hector mine sites (names in _GHT), logarithmic fit with log function starting 10-days before the earthquake included in the fit.
JPL_96_0210.vel	Same style of file as COM_96_0210.vel except only JPL STACOV files used in the analysis
JPL_96_0210_RS.vel	Same style of file as COM_96_0210_RS.vel except only JPL STACOV files used in the analysis
JPL_Proc.vel	JPL Velocity estimates from table2.txt rotated from ITRF00 to North America Fixed using the ITRF 2000 NOAM rotation pole.
SIO_96_0210.vel	Same style of file as COM_96_0210.vel except only SIO h-files used in the analysis
SIO_96_0210_RS.vel	Same style of file as COM_96_0210_RS.vel except only SIO h-files used in the analysis
SIO_Proc.vel	SIO Velocity estimates from refinedModelVelocities.cgi rotated into a NOAM frame using ITRF 2000 pole. Latitude and longitudes added the file so that results may be plotted. These velocities are the current velocities of the GPS sites (see documentation on the SOPAC web site. Some standards deviations have been replaced with non-zero values at sites where zero sigma is reported in the SOPAC file.
mpb_com	Time series from the combined analysis. File names are of the form mb_NNNN_XXX.datY where NNNN is the four character code for the site, XXX is the extent that indicates post-earthquake and/or jumps in the time seires, and Y is 1 for North, 2 for East and 3 for Height components. Time is in years. Time series are in a NOAM frame
mpb_jpl	Similar to mpb_com but from processing JPL STACOV files
jpl_fil	Time series from the JPL web site series.html with results ftp’d from sideshow.jpl.nasa.gov. The mb_xxxx files are in the tsview format. The original files from the JPL are also included in the directoty. NOTE: These times are in the ITRF 2000 frame. The itrf00_noami00.dvel file linked below gives the offsets in velocities between ITRF 2000 and NOAM for sites in the California region.
mpb_sio	Similar to mpb_com but from processing SIO h-files
sio_fil	Times series from SIO web site obtained with wget from refinedTimeseriesListing.cgi. These files have been converted to the mb_xxxx files and the time series limited to 1996 to Oct 2002 to match the JPL time series availability. The original SIO files are also included in the directory. These time series are also in the ITRF 2000 frame.
itrf00_noami00.dvel	File containing the difference in North and East velocities between ITRF 2000 and NOAM for sites in California.
CCVelV.xy	Simple file of longitudes and latitudes of the California coast appropriate for use with Features option in velview. NOTE: The features option does not currently work in the executable version of velview; velview must by run from within Matlab to use this feature.
CF3VelV.xy	Simple file of longitudes and latitudes of the California Faults appropriate for use with Features option in velview
CEQ2VelV.xy	Simple file of longitudes and latitudes of the California earthquakes since 1996 with magnitude >2 appropriate for use with Features option in velview (use the brown dots load, Dbrown). Data obtained from the USGS NEIC
CEQ3VelV.xy	Earthquakes >3 magnitude since 1996
CEQ4VelV.xy	Earthquakes >4 magnitude since 1996

Earthquake names:

LA – Landers, LB – close in region to Landers with extended postseismic relaxation time, NR – Northridge, NA – After shocks of Northridge; HT – Hector mine.

The combined motion field was generated by combining the JPL and SIO motion models generated from the data sets described below using GLOBK. There are still some issues to be resolved in the combined field.

Issues to be resolved:

Issue	Resolution
Antenna heights and types	As noted below there are a number of antenna height differences between the JPL and SIO analyses that have been fixed in the current analysis for most sites. There are some remaining sites with non-Dorne Margolin antennas (antenna type is not recorded in the JPL STACOV files) for which this correction might not be done correctly.
Discontinuities	There are still some discontinuities in the times series that need to be accounted for. These corrections will be made when then next batch of JPL STACOV files is available.

Motion Maps:

Plotted below are below:

SCIGN Velocity field in North America fixed frame. Red vectors are from the GLOBK analysis of sites showing small non-secular motions; Blue vectors are from time series analyses of all sites. Only sites with velocity sigmas less than 1 mm/yr are shown. Error ellipses and site names are not shown.

SCIGN combined velocity field with the average motions in the Los Angeles basin set to zero. NOTE: Much of the motion seen is due to strain accumulation on the San Andreas Fault,

Basic Data Sets

The “data sets” used in this analysis were:

1. From JPL: loosely constrained “stacov” files that contain station position estimates, covariance matrix, and limited information about antenna heights. For each day, the stacov files contain all the sites used on that day obtained by merging results from up to six network. The JPL merged stacov files were used in this analysis. The files made available to the SCIGN analysis group ran from Jan 1, 1996 to October 12, 2002. The sites include were in the box between longitudes 239-250 and latitudes 28-39 degrees. All the SCIGN sites and some, but not all, of the BARGEN sites this box were included. The stacov files are available at ftp://sideshow.jpl.nasa.gov/pub/mbh/stacov/loosefinal.

2. From SOPAC: loosely constrained GAMIT “h-files” that contain station positions, satellite orbital parameter, and Earth Orientation parameter estimates; covariance matrix and antenna type and height information. For each day, the h-files are divided into networks with up to 15 networks per day containing about 450 sites distributed around the world. The SCIGN group of stations is represented in up to 6 networks again with a common group of stations tying the networks together. In this case, the common sites are disturbed across North America. These files are available for data collected since 1990 and updated each week with about a 3-week delay from real-time (ftp://garner,ucsd.edu/pub/hfiles). For this analysis, we used only sites in the latitude/longitude box and time interval given above.

Analysis methods

The approach in the analysis method was to treat the data from both groups in the same way using the GLOBK software, which is designed to combined position estimates and covariance matrices using a Kalman filter approach.

For the JPL stacov files we changed the antennas at 46 sites to make them the same as the heights in SOPAC site database. For the Dorne Margolin antennas this change can be made exactly. For other antenna types we had to assume that phase center model in the JPL analysis was the same as the SOPAC analysis and that JPL had not applied horizontal position changes for these antennas (such as the TRM14532.00 antenna at MATH) because only an LC height is given in the stacov files with no information about the models used for either the antenna type or its characteristics as the L1 and L2 frequencies.

Both sets of solution files were combined into daily files that contained all the stations processed by each group. For the SOPAC h-files, this combination used all the sites in all the networks and the orbital parameters were included in the combination. From these combined files, only the sites in the SCIGN latitude/longitude box were used for analysis. (Since the representation of the solution is in covariance matrix space, the step simply requires creating a covariance matrix that contains the rows and columns of the desired sites.) The orbits were loosely constrained in these combinations.

The combinations resulted in 2472 daily files for SOPAC (the expected number) and 2456 files for JPL.

For this preliminary comparison the velocity field and times series from these daily files were constructed in two steps: “Good site combination” and “Remaining Sites Analysis.” The production of the combined velocity is being carried out in stages. In the initial analysis, http://bowie.mit.edu/~tah/SCIGN/SCIGN_AC09_2002.html, we used 86 “good” sites that had more than 500 days of data and horizontal RMS scatters less than 1 mm. With this selection of sites there were problems straddling the discontinuities in the time series due to the Hector Mine earthquake. In the current analysis, we have been less restrictive in the selection of the core sites and we had added an additional 6 months of data (recently made available by JPL).

“Good Site Combination”

Because a number of SCIGN sites have systematic temporal variations and in some cases poor data quality during some intervals, we selected 264 “sites” in the SCIGN box that had more than 500-days of data and an RMS scatter in the horizontal position estimates of less than 2 mm. This selection was based on previous on the preliminary SCIGN analysis of September 2002. The 264 “sites” represent 195 distinct stations because in our treatment of sites, we rename stations after earthquakes and if there are non-modeled jumps in time series (often due to raydome and/or antenna changes).

In separate analyses, the JPL and SOPAC files were combined with GLOBK to estimate the position and velocities of the sites. After the Hector mine earthquake (1999/10/16), we renamed all sites within 250km of the epicenter (34.2752N –118.5962E) and included random walk process noise in the filter for 160 days after the earthquake. The minimum process noise was 0.1 mm²/day with it increasing for sites closer to the epicenter. We also renamed sites that had unexplained jumps. This information is contained in the globk earthquake file labn.eq. A notable change to this file from the September analysis is that we now allow jumps when antennas are changed from the injected Ashtech Choke ring antennas (ASH700936E_C) to the milled choke ring antennas (ASH701945B_M). This change in antenna type causes several millimeters of horizontal position change and does not appear to be the same at all locations.

To keep the velocity sigmas realistic we added random walk process noise to all sites of 0.3 mm²/yr for both analyses. The effect of the process noise is to limit the velocity sigmas to no less than 0.17 mm/yr. The combination was performed with loose constraints on the positions and velocities. The final loose solutions were rotated and translated onto a common, nominally North America fixed frame derived from the SCEC Crustal Motion Model Version 3.0. (Since the JPL stacov files were limited to sites in Southern California, it was not possible to use site located on stable North America).

The statistics of the differences are:

For the 264 site names in this analysis:

Horizontal components only (2-D): WRMS difference 0.71 mm/yr, NRMS 2.56

3-D components: WRMS 0.86 mm/yr, NRMS 2.66

If we estimate a rotation and translation between the two fields, these difference reduce to:

2-D WRMS 0.60 mm/yr; NRMS 2.18 and

3-D WRMS 0.74 mm/yr, NRMS 2.28.

(WRMS is Weighted Root-Mean-Square, and NRMS is Normalized Root-Mean-Square and is equal to (square root of Chi squared per degree of freedom).

“Remaining Sites Analysis”

The second analysis adopted a reference frame based on the 264-sites from the above analysis. Time series were generated in this analysis by rotating and translating each daily combined file on the reference frame. Linear trends were fit to North, East and Height components from the time series.

The overall statistics of these time series are for JPL for 448 site names:

NORTH : 50% < 1.2 (mm) 70% < 1.5 (mm) 95% < 3.8 (mm)

NORTH : Mean (mm) : 1.6 Sigma (mm) : 1.2 Stations: 448

EAST : 50% < 1.4 (mm) 70% < 1.8 (mm) 95% < 3.1 (mm)

EAST : Mean (mm) : 1.7 Sigma (mm) : 1.1 Stations: 448

UP : 50% < 4.2 (mm) 70% < 4.8 (mm) 95% < 9.1 (mm)

UP : Mean (mm) : 4.9 Sigma (mm) : 3.3 Stations: 448

For SIO:

Statistics are for 508 sites (not individual locations):

NORTH : 50% < 0.8 (mm) 70% < 1.1 (mm) 95% < 2.4 (mm)

NORTH : Mean (mm) : 1.0 Sigma (mm) : 0.8 Stations: 508

EAST : 50% < 0.8 (mm) 70% < 1.0 (mm) 95% < 2.5 (mm)

EAST : Mean (mm) : 1.1 Sigma (mm) : 0.8 Stations: 508

UP : 50% < 2.9 (mm) 70% < 3.6 (mm) 95% < 6.9 (mm)

UP : Mean (mm) : 3.6 Sigma (mm) : 2.7 Stations: 508

Analysis of the Times series determined velocities

The standard deviations of the velocities from the times series are too small because of the white noise assumption which can be clearly seen to be violated by many (and on close examination all) time series. The comparisons below use the standard deviation for weighting the root-mean-square calculation and sites whose velocities have very small standard deviations will tend to dominate the statistics.

Between the two time series based velocities fields there are 440 common site names (separate velocities before and after earthquakes). The statistics of the comparison of the two fields are:

For 440 sites:

2-D WRMS 0.60 mm/yr; NRMS 5.59

3-D WRMS 0.76 mm/yr; NRMS 5.93

If a rotation and translation between the two fields is estimated, the statistics are:

2-D WRMS 0.56 mm/yr; NRMS 5.27

3-D WRMS 0.71 mm/yr; NRMS 5.52

The NRMS values are high because of the white noise statistics assumption.