Effect on Data

The effect of the EOP errors on an individual observation depends on the quality of the predicts used and on the geometry of the observation. The errors are dominated by the error in UT1-UTC which has an effect equivalent to an error in the RA of the source. The magnitude of the equivalent position error (UT1-UTC is measured in milliseconds of time which has to be multiplied by 15 to get milliarcseconds) is typically about 10 mas before mid 2004 and 15 mas (with more observations that are significantly worse) after that. The worst cases can be around 4 times these numbers. The X and Y pole position errors are also equivalent to a source position shift at any one instant, but the direction of that equivalent shift rotates through the day. The magnitude is typically around 20% of the UT effect. For details on exactly how the EOP values affect calculated delays, consult the references given at the end of this document.

When phase referencing, the self-calibration done on the calibrator will remove any effect of the EOP error at the position of the calibrator. The phase referencing will then apply that same phase correction to the target source. But the phase correction actually needed is a sine function of hour angle and declination so the correction from the calibrator is not quite right for the target source. The error is, to first order, the total magnitude of the effect times the source-calibrator offset in radians. The exact effect is more complicated and won't be described here.

The model errors resulting from poor EOP appear in the form of delay errors and corresponding phase offsets and slopes. These result in the position errors discussed above. The delay errors are typically a few ns and introduce phase slopes that vary between sources. Spectral line observations that depend on a calibrator to set the delay, and hence the phase relationships between channels, will be affected by such a phase slope. For an 8 MHz bandwidth, a 4 ns delay offset (in the range seen) corresponds to a phase slope of 11.5 degrees across the band. For wider bandwidths, including using multiple basebands, the effect will be larger. For some astrometric observations, such a slope may be significant.

To help users check the EOP errors for their observations, tables that contain the EOP values used and the offset from the best values known as of 2005 Sept. 7 have been placed at ftp://ftp.aoc.nrao.edu/pub/staff/cwalker/VLBA. There are separate files for each year between 1998 and 2005. There are other tables of the values used on the correlator, good values from a usno_finals.erp file, and differences between these two, plus plots of the differences, at http://www.aoc.nrao.edu/%7Evdhawan/eop.html. A user may also determine the EOP values used in an observation by looking at the CT table attached to the AIPS data set or by looking at the UTC Table near the end of the job scripts in the NRAO archive at http://www.vlba.nrao.edu/astro/VOBS/astronomy/. Good values for the EOP parameters may be found by following links from http://www.iers.org/iers/products/eop/. Perhaps the best values to use for VLBI are in the previously mention file http://gemini.gsfc.nasa.gov/solve_save/usno_finals.erp. That file is updated daily. It contains the latest information from the IERS site but adjusted by a small offset and drift to be a better match to the source and station catalogs used at the VLBI correlators.

One simple step that users can take to check their data, if they scheduled according to recommendations, is to see how well the phase referencing worked on their phase check source. Recall that it is recommended that a second calibrator, near the primary calibrator, be observed occasionally. That calibrator is used to check the quality of the phase referencing by checking how close a position derived from phase referencing comes to the known astrometric position. Information about the quality of the phase referencing conditions also comes from checking the dynamic range of the phase referenced image of the secondary calibrator. This is a good check, but not perfect because the impact of the EOP error depends on the orientation of the offset between calibrator and target which is likely to be different for the calibration check source and the object being studied.

It is not difficult to fix the effects of the use of incorrect EOP values. The effects on data phase are large enough to damage phase referencing, but they are nowhere near large enough to cause loss or degradation of fringes. At the time the problem was found, there was no available method in AIPS for correcting the data. Since then, an option has been added to CLCOR to allow the corrections to be made. Details are discussed in Section 7 and in outline form above in Section 2.

Figure 1: Plot of the offsets between the EOP parameters used in correlator jobs and reference good EOP values from a usno_finals.erp file. Data are shown for times since 1998. Vertical bars mark the year boundaries with the last partial year being 2005. The correlator jobs began to include poor EOP in May 2003 and were corrected in August 2005 just before the last data shown here. Longer range predicted EOP values started being used in June 2004, accounting for the worsening of the offsets at that time.

Figure 2: Histograms of the same data shown in Figure 1 for different time ranges. The horizontal axis is the square root of the sum of the squares of the different offsets, which is meant to be a measure of the total error even though UT1-UTC is not orthogonal to the other terms. Note the change of the horizontal scales between plots. The top plot shows the offsets seen while using the rapid service values. Most are better than 1 mas total error, while a few were worse. After May 2003, the number of projects with poor EOP increased dramatically, and it got even worse after June 2004.

For an overview of the effect on VLBA projects, Figure 1 shows a plot of the offsets between the EOP values used and the values in a usno_finals.erp file as a function of date. The vertical bars are the year boundaries. Prior to early 2003, most observations had good values. The exceptions are probably jobs prepared sufficiently close to the observe date that rapid service values had not yet been given for all days with EOP included in the job scripts. Around May 2003, a much larger portion of observations had significant offsets. There are still some good values because the correlator still used rapid service values for 2 days a week. After June 2004, when longer range predicts started being loaded to the data base and used, the errors got significantly worse. Figure 2 shows the same data as Figure 1, but in histogram form. There are 3 histograms for the different time periods.

Since the effect on any given data set depends on details of the observation setup and geometry and on the EOP errors encountered, probably the easiest way to test the effect is to run CLCOR as described in other sections and examine the magnitude of the changes made.