TCCON > Sites > Białystok > Timing errors

Timing errors

Table of contents
  1. 1. Summary of derived tcorr

Notes on diagnosing and correcting timing errors at Bialystok

 

These are apparent in the xAIR time series. Why?

 

At the end of the time series, the NTP server was not syncing, but why are there errors earlier? The correction applied in late 2013 and early 2014 is as follows:

 

          if(jd_meas.ge.jd_corr(1).and.jd_meas.lt.jd_corr(2)) then
              tcorr = 179.5*(jd_meas-jd_corr(1))
     & /(jd_corr(2)-jd_corr(1))     
c              write(*,*) 'tcorr', tcorr
          endif
        

where jd_corr(1) is Aug 4, 2013 and jd_corr(2) is Aug 19, 2014. 

This assumes a linear time drift from when the NTP sync failed until it was discovered, with a time lag of 179.5 seconds.

 

The timing of this doesn't seem quite right, however, as the xAIR shows diel patterns from Aug 1 onwards.


 

E.g. there are clear diel patterns from mid April 22, 2009 for the following week or so. Then at the end of 2009 and into 2010 there are also diel patterns (less distinct, so possibly smaller magnitude timing error).

At least for April 22 - early May, these are not obvious from the NTP logs, which appear to indicate that the time was updated. The drift starts at around 11.20am on April 22. We then have hte message "20 May 07:10:14 ntpdate.exe[1356]: step time server 134.34.3.18 offset 13.169152 sec" at the start of the next log (though there are many entries at the end of the previous with dates between April 22 and May 20).

The fact that this starts midway through a day means that the method of using the julian day to check for tcorrs will not work appropriately, and we need to work with decimal days instead. :(

 

Testing the tcorr on April 22, with a test value of 13seconds. To run fits on April 22 only takes about 15 minutes (O2 window limited) when separating into the individual windows.

* this seems to be in the correct direction, but by far the wrong magnitude. we try next with 130 seconds

* 130 seconds seems much closer to reasonable, but still not perfect (not that I expected it would be). trying 150. trial and error approach at present...

* 150 not different from 130... 140?

* 140 is marginally better than both 130 and 150, but still not great (xair varies by ~1% compared to the AM). Where to from here?

* 200 is no better, but not worse than 130 or 150. I will include April 23 for a full day with time error

* some idiot was plotting 130 twice. 200 looks not too bad, but maybe just a little large (i.e. PM goes up now). Try 180.

* so an xAIR error of ~2% @ 70degrees seems to be about 180 seconds at Bialystok. This gives us something to work from...

* 180 seconds seems reasonable for this time frame. I will now fit and check for the full period that this time error appears to exist

 

*For Dec 14, 2009 to Feb 9, 2010, I use Jan 22, 2010 as a test day. 180 seconds is too large, and over-corrects. I would guess about 60-80 seconds is closer...

* 80s seems pretty close, but maybe slightly too much. I will try 60 (seems somehow reasonable to believe it might be off by 1 minute, rather than 70 or 80s, but why, I don't know...)

* 60 looks worse than 80, so maybe try 70 and then choose. 70 and 80 look about equally good. going with 70

 

* next stop, Jun 01, 2010 - Jul 27, 2010

* similar to late 2009? starting with 70s

* not bad, but could be better (a little larger). 100s

* 100s looks good

 

* Update ... for Dec 2009 - Jan 2010, the timing error appears to drift. The correction is perfect at Jan 22, but too big beforehand and too small afterwards. So we need to derive a time-dependent correction. I will look at the earliest and latest days and assume a linear drift in between times.

 

* Late August 2009 - 30s correction seems too large. Will try 15. However, some days seem to show a bit of symmetry (airmass dependence?), so we should aim for the most symmetrical, I guess. On a plot of xAIR vs SZA, the 15sec correction seems flattest/most symmetrical, but there's not much in it.

 

* at current state, 2009 looks okay, but I am fitting the entire year to be sure

 

* mid 2010 (jun-jul) - current correction seems to overcorrect early and undercorrect late, so trying a larger time dependent drift (50 + 2*day). Assessment complicated by lack of symmetric days (around noon), meaning that it could simply be airmass dependence that is showing up

 

* 2009 corrections look good - no obvious symmetry in xAIR vs SZA plots.

 

* 2010 between Feb 9 and Jun 1 still seems to have some timing problems. Trying 50s... Too big by ~ factor 2. 30s... looks good.

Summary of derived tcorr

  • April 22, 2009 @ 11:20 - end April 2009: 180s  (update: this seems okay as a constant tcorr)
  • Aug 21, 2009 - Aug 31, 2009: 15s
  • Sep 2009 - Nov 2009: 30s
  • Dec 14, 2009 - Feb 9 2010: ~70s (time-dependent, with drift of about 1s/day)
  • Feb 9, 2010 - Jun 1, 2010: 30s
  • Jun 1, 2010 - Jul 27, 2010: ~100s (as above, time-dependent, with drift of about 1s/day)
  • October 2010 - March 2011: 40s (fairly time constant)
Tag page