PAGE UNDER CONSTRUCTION
Take everything herein with a grain of salt till I am done editing and this message disappears.
SCIGN Analysis 3.0.0

The initial posting of these results covers the period from January 01, 1996 to August 27 2001.

TOP
Repeatabilities
Strategies and Methods
Differences between this and previous JPL analysis
Format of data files
Calculating Offsets
SCIGN_3.0.0 position series data

Repeatabilities

Strategies and methods

The RINEX data from each station on each day was first processed using precise point positioning methods. In this technique, precise GPS orbits, GPS clocks, and earth orientation information derived from a previous analysis of  global GPS data (our FLINN processing), are held fixed, and the station position, white noise station clock,  and random walk zenith troposphere delay and gradient are estimated using a 15 degree elevation cutoff.  The GPS station antenna phase center was assumed to be an invariant point. In other words, no phase center maps were used. Ocean loading was modeled, for almost all stations. A few stations which did not have their own ocean loading coefficeints "borrowed" coefficents from another station. A table showing where the tidal coefficients came from is provided here. For those of you who are fluent in  GIPSY-speak, this step is done by point_rnx, which in turn uses xt-gipsy with the following namelist inputs:

TROPDRIFT = 1.7d-7
SSTCH= 'TRPAZCOSDSSC', 'TRPAZSINDSSC',
SMTAU = 1*'RANDOMWALK', 1*'RANDOMWALK' !trop
SPSIG = 1*5.0d-9, 1*5.0d-9 !trop
SDELT = 1*'00:05:00.000000', 1*'00:05:00.000000' !trop

ELMINSTA = 15.0

After a suitable quantity of data has been processed through ambiguity resolution, the data is edited to delete obvious outliers.  The 12 common stations are then combined in a least squares process to estimate their positions and  velocities, again in ITRF97.  The solution for the positions and velocities of these 12 stations then becomes an interim regional definition of the ITRF97 reference frame. These 12 stations are chosen to be distributed around the periphery of the network. It is imperative that ALL of the data for these 12 stations be used to determine their positions and velocities, or the interim regional ITRF97 frame will not be congruent with the global definition. Each solution from the ambiguity resolution run is then transformed with a 7 parameter transformation into the interim regional ITRF97 reference frame.

The resulting solutions are then combined in a least squares process to estimate station positions and velocities for all the sites. This combined solution then becomes the final regional ITRF97 reference frame, using all the data from all the stations to help define it. A final set of 7 transformation parameters is determined for each individual solution to put it in the final regional ITRF97. Time series of station positions and daily repeatabilities are calculated with reference to this combined solution.

The result is a time series of station positions which has a noise level commensurate with the baseline precision. The noise in a baseline series is about sqrt(2) times the noise in a single station series. The baseline precision has been improved over the original point positioning through ambiguity resolution.

Another way to think of this is that ambiguity resolution allows you to ask a different question of the data. Instead of asking "where is this station?" you are better off asking 2 questions: "1) Where is the centroid of the network? and 2) What is the baseline from the centroid to this station?"  Ambiguity resolution helps refine the answer to question 2, because it adds very strong information about the baselines in the network,  and the  regional ITRF97 definition described above helps to answer question 1.

How are these series different from the JPL SCIGN 1.0 and 2.0 analysis?

1. The 1.0 analysis used precise point positioning, leaving the phase ambiguities as real values. The 2.0 and 3.0 analysis also do precise point positioning but then go on to resolve the double difference phase ambiguities to integer values.
2. The 1.0 analysis is expressed in a global definition of ITRF97. The 2.0 and 3.0 analysis use a regional definition of ITRF97. It is important to realize that this regional ITRF97 really is ITRF97, not an arbitrarily defined, unconnected local reference frame. As such, there is no improvement in the estimate of the steady-state station velocities. The velocities should be identical to what is obtained from the global ITRF97 analysis.  The daily scatter is reduced, allowing examination of  smaller time-variant signals.
3. The 1.0 analysis had daily station coordinate repeatabilities of 3.7 mm in latitude,  5.5 mm in longitude, and 10.3 mm in the vertical. The 2.0 analysis has daily repeatabilities of 1.2 mm in latitude, 1.3 mm in longitude, and 4.4  mm in the vertical.
4. The 1.0 analysis uses a large amount of hand editing in the time series analysis. Almost all of the time series analysis has been automated in the 2.0 and 3.0 analysis.
5. There are relatively minor differences in the point positioning analysis, though they are done with completely independent scripts. For those fluent in GIPSY, the 1.0 analysis was done using p3. The 2.0 and 3.0 analysis was done using point_amb (which subsumes point_rnx and xt-gipsy) and amb_scomb

Format

A simple perl script to parse these files and to write various pieces of the information to stdout is available here
Another simple perl script which uses gnup and gnuplot to plot the time series of a station's position is available here.
Note:  These programs run on the development HPs but not the production HPs yet. (July 27 2000)

The perl packages needed to run this script are:

1. XML::Simple.pm
2. XML::Parser (includes (XML::Parser:Expat)


In addition, people using perl 5.004 will need:
3. File::Spec


and some poor souls may need:
4. Storable.pm

All the positions and velocities have been propagated to the epoch date of the whole file. (given in $station->{epoch}). Each position and velocity given for a station also has a time of validity which is when that velocity should start to be effective. (given in $station->{location}->[0]->{float_year} for the first one)

Calculating offsets

The best way is probably to average a few days before and a few days after the event and subtract the two averages. The time span of the averages should be as long as possible such that the anticipated motion of the station during the time between midpoints of the averaging intervals is negligible.

Another method is to fit a station velocity to the time series of position of the station (possibly having different velocities before and after, as might be appropriate for an earthquake, or perhaps forcing the velocity to be the same before and after, as might be appropriate for an antenna swap). The fitted positions and velocities before and after are then projected to the time of the event and subtracted to yield the offset. This method is critically dependent on getting the correct velocity of the station both before and after the event. As such, it is subject to the vagaries of bad data, and the exact timing of other events which demark the sections of data used to determine the station velocity. Hence I regard it as less robust than the simple averaging procedure above.

John Langbein has also pointed out that the uncertainty in the velocity will depend on the noise model assumed. At present, White Noies, Flicker noise, and Random Walk are all contenders for describing all or part of the noise budget in GPS time series.

The user should decide on a method of calculating any offsets of interest, and apply it to the raw data presented in these data files.

Data

TOP
Repeatabilities
Strategies and Methods
Differences between this and previous JPL analysis
Format of data files
Calculating Offsets
SCIGN_3.0.0 position series data
SCIGN_3.0.0 position series plots
SCIGN_3.0.0 tropospheric delay data
SCIGN_3.0.0 tropospheric delay plots