Investigation of the differences between gsfint02.eopi and gsfint01.eopi series

Investigation of the differences between gsfint02.eopi and gsfint01.eopi series.

The problem

It was noticed that the differences between the two UT1 series, gsfint02 and gsfint01, are larger than expected: 2 nrad (30 microsec of time). They show clear systematic behavior. Results from the experiments on the baselines NRAO20/WETTZELL systematically differ from the experiments on the baselines KOKEE/WETTZELL.

Investigation

Comparison between the estimates of UT1 between the old series gsfint01.eopi and gsf2000a.eops derived from the 24 hour EOP experiments shows wrms difference 28 musec with sqrt(chi/ndg) = 1.83 over 120 points, while the differences in UT1 between the new series gsfint02.eopi and gsf2001a.eops shows wrms 26 musec with sqrt(chi/ndg) = 1.83 over 190 points. Conclusion: the new UT1 series has slightly better accuracy than the old one, but this difference is not statically significant.

Comparison with solutions provided by other VLBI Analysis centers shows that the new series gsfint02.eopi agrees worse with other submissions than the previous series gsfint02.eopi .

Analysis of Solve control files for gsfint01 and for for gsfint02 reveals only two differences:

The differences in apriori positions of stations and sources.
Solution gsfint02 had more sessions.

Certainly the latter difference cannot contribute to the differences in UT1, since, unlike 24 hours VLBI sessions, our group analyzes the Intensives independently.

I made solution gsfint02b with exactly the same apriori as gsfint01. The plot of the differences in UT1 shows one order of magnitude less differences.

Impasse?

But there is another source of differences. Unfortunately, our group doesn't have resources to maintain our own near-real service of EOP prediction and uses the USNO's EOP product, namely finals.all as the apriori series for processing Intensives.

This product is continuously updated by USNO. Our service browses this file every night. Therefore, two solutions, gsfint01 and gsfint02, have different apriori pole coordinates and UT1.

I recovered the old version of our a priori file with the end date 2000.11.09, the version which was used for testing gsfint01 solution, and made a new solution gsfint02a which differs from gsfint02 by only using the old version of of our a priori file. The differences in the estimates of UT1 between gsfint02 and gsfint02a reveals exactly the same pattern as the differences between gsfint02 and gsfint01 UT1 series ( The differences at the right edge of the plot are due to the fact that the extrapolated EOP were used in gsfint02a solution for those dates ).
The problem is solved.

The was a note from USNO with warning that the bug was discovered in the old series around 01 February 2001. This bug affected all previous analyses which relied on USNO finals eop series, including our previous solution gsfint01.eopi . Other possible factor is the change in the procedure of transormation of the origianl USNO eop fiel to our a priori file.

Conclusions

The difference in UT1 gsfint02 minus gsfint01 is due to the differences in our a priori EOP files.
Series gsfint02.eopi are better than gsfint01.eopi
We should not have illusions that comparison of EOP series with those obtained by other AC's will always help us to discover the problem.
Closer does not mean better!

I am thankfull for Jim Ray for noticing the problem and setting alarm.

A priori EOP series used for gsfint02a.eopi soltions usno_finals_old.erp.
A priori EOP series used for gsfint01a.eopi soltions usno_finals.erp.

Back to Leonid Petrov's discussion page.

Last update: 07-MAR-2001 12:34:45