[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: long term variation and instrumental drift
Chris Lloyd wrote:
>Long term changes can be caused by aging of the filters but this should
>result in a very slow evolution of the transformations coefficients. I'm
>not sure if I understood correctly about the transformations coefficients
>being re-determined every night - this is not a good idea.
It can be ok if the quality of the re-determination is reasonable.
Fixing the transformation coefficients, when you combine the results
from several nights, reduces the mean error and therefore gives you
better transformation accuracy, provided that the average coefficients
are good. Sometimes you get systematics, just like in flats, and then
using the systematically incorrect coefficients makes all the processed
data have systematic errors. For that reason, I tend to form averages
per run and apply them, but do lots of plots of seasonal trends and
also check the deviation of the determination on a nightly basis.
Doing automated processing is fine, as long as you perform quality assurance.
The problem with transformation coefficients is that they are a function
of color. For example, you might have
V = vins + zeropoint + slope*(V-I) +extinction
The zeropoint is determined by Michael's use of Tycho2 stars in the
particular field; depending on the spectral type of those stars and
their number, this is probably good to +/- 0.05mag or better. The
use of Tycho2 stars within the field also removes the constant extinction
term; it gets folded into the zeropoint (though there will be a
gradient across the field that becomes important at high airmass).
The slope, however, is what Michael says he redetermines nightly (or at
least processing-runly). If it is a photometric night, the slope can
probably be calculated reasonably accurately. If it is a non-photometric
night, I worry a bit about the quality of this determination.
It becomes important when then applying the transformation 'slope' to
actual data. This is especially true for the red stars (likely variables,
and so can be the majority of variables found). If (V-I) is 2, for
example, then using a slope of 1.05 vs. a slope of 1.10 will change
the derived magnitude by 0.1mag, much larger than the Poisson error.
In my mind, such slope or zeropoint errors are likely culprits if you
see a night-to-night jump in the magnitude for a given object.
Tom, this is just a processing step that needs to be checked
thoroughly at this 'experimental' stage. I am not asking you to retake
data or to do anything that any other photometrist wouldn't do with
their data before publication.