The receiver stores the recorded time, pseudorange and phase data for each tracked satellite in a non-volatile RAM PCMCIA card. In standard operating procedure, the receiver logs this data to the flashcard once every thirty seconds. A typical site is monitored for three full days.

When the receiver is brought back to campus, its serial port is hooked up to a personal computer. Using a terminal program, the data are offloaded in a proprietary binary format defined by the manufacturers of the receiver. There are however, several different commercial manufacturers of GPS geodetic receivers, each with their own internal data format. In order to share these data with other researchers and use standard software for its subsequent processing, the next step will be to translate it into an internationally agreed-on standard format, the so-called RINEX text format. However, in its present, compressed binary (.cb) format, a single day's data occupies about 420 kb.

When the data are stored safely on the personal computer, they are transferred to the UNIX server where the data processing will be done. Ideally, there is at least one backup copy of the data, in addition to the copy that is being interpreted on the UNIX machine. It is extremely frustrating to have three days of effort suddenly disappear into a flurry of random electrons.

Once transferred to the UNIX workstation, a translation program is run which converts the 420 kb raw binary file into about 1.8 Mb of ASCII text RINEX file. It is then ready for processing by GIPSY, a set of precision GPS processing tools written at Jet Propulsion Laboratory (JPL Link?) and comprising about 100,000 lines of C and FORTRAN source code.

In addition to the RINEX data for our own site(s), we also obtain from public archives the GPS data from other permanently operating sites in the region. (Relative differential positions determined between nearby sites are generally more accurate than absolute positions determined for isolated sites.) Before we can proceed with processing, we must also obtain from JPL (or another GPS processing institution) their precise solutions for the satellite orbit ephemerides and clock corrections. Although we could in principle solve for these unknowns ourselves, fixing them at known values in our solutions improves the quality and strength of the remaining unknown parameters.

The first step of the GIPSY processing is "data editing". Once done laboriously by hand by graduate student labor in the dark ages of the '80's, this task is now accomplished in software using pattern recognition. It is necessary to "clean" the data of sprious jumps in in phase or "cycle slips" caused by hardware glitches or momentary interruptions of signal.

Once this step is complete, the GIPSY software assembles a huge simultaneous-equations least squares problem, in which the unknowns are such things as site position, receiver clock error, atmospheric water vapor content, etc. The data are thousands of values of phase, pseudorange, etc. taken over a 24-hour period. The software then solves this system of equations to obtain a "best estimate" and uncertainty for each of the unknown parameters. Among these parameters is what we really want to know -- the precise three-dimensional location in space of the receiver phase center(s).

The above description is an oversimplification, as the data are actually processed iteratively, with bad data judiciously edited to improve the chi-squared quality of fit of the "raw" data to the GPS model. In practice, because of unmodeled error sources, the final chi-squared value is not unity, but the "formal" uncertainty estimates emerging from the analysis for any given 24-hour period are usually within a fact of 2 to 4 for being realistic.

Once projected into a conventional east-north-up coordinate frame, the positions calculated for various sites are compared with positions at the same sites measured months or years earlier. From these, velocity vector components are estimated. (In unusual circumstances, such as following major earthquakes, motions that are non-constant in time can be detected.) By browsing examples of these velocity trends found on this web site, you'll notice that the statistically significant velocities are mostly horizontal. Vertical velocities tend to be much smaller, and the vertical measurements also have inherently larger uncertainties.

Vectors in this case represent a velocity, but a velocity relative to what? In this study, vectors are measured relative to the position of the GOLD site. The GOLD/GOL2 site is one of world's oldest continuously recording GPS sites, and is located at the Goldstone Deep Space Network tracking complex in the Mojave desert about 100 miles northeast of the Los Angeles basin. In order to gauge the movement of the earth locally, it is necessary to have a "stationary" reference point. Ideally, this site would be as far away from the geologic "action" as possible. Unfortunately, the as the reference site gets farther away, more error is introduced into the measured distance to that site. For example, the distance to a site in New Mexico might have an error measured in centimeters, whilst the distance to GOLD might have an error measured in millimeters. If the error in this distance is on the same scale as the movement of the earth that is monitored, it is difficult to get a meaningful vector relative to the reference site. GOLD is a good compromise between proximity to the studied area and geological isolation from that area.