5.2. Data Editing

Your solutions will be corrupted if you include data that are statistical outliers. For a time series from independent h-files, an error in one component of one site’s coordinates will not affect any other site (and may have minimal effect on the other components), but as soon as you combine h-files with common parameters, whether multiple networks on a single day or the results from more than one day, the different estimates of the common parameter will clash, raising chi-square and distorting the solution. There are two ways you can remove outliers: using the rename command of the eq_file and using the sig_neu command directly in the globk command file. In the rename command, if you designate the extent as _XCL the site will always be excluded; with _XPS the site will be excluded from any solution of more than one day but will still appear in daily time series; e.g.

rename  MIT  algo_gps  algo_xcl  1997  5 14  0  0  1997  5 17 24  0

will remove from the solution the observations of ALGO in any h-files containing the characters MIT between 14 and 17 May 1997. Note that the date span specified must completely encompass the start time of the first h-file and the end time of the last h-file. The h-file selector entry and/or the dates may be omitted for a more expansive exclusion. With the sig_neu command you can effectively remove the effect of an error in a particular component by adding sufficient noise that it has negligible weight, e.g.

sig_neu  MIT  algo_gps  0  0  .5  1997  5 14  0  0  1997  5 18  0  0

would reduce the effect of an error in height by adding \(0.25 \text{m}^2\) in quadrature to the variance of the height component as estimated from the h-files specified.

Which of these approaches you choose is a matter of taste and could depend on the tools you use for editing. For continuous data, when you can afford to remove many outliers without weakening the solution, using the nsigma option of the tsfit program (see Chapter 4 of the GLOBK Reference Manual) to automatically generate _XPS commands, or the MATLAB-based interactive program tsview to generate _XPS commands with the click of a mouse, are quite convenient. For survey-mode date, when every observation counts and you often want to down-weight heights but not horizontal components or only one horizontal component, sig_neu commands are more convenient. The GAMIT utility grw can be used to generate sig_neu commands with only a few key-strokes.